Identification and Management of Epidemic Hazard Areas for Urban Sustainability: A Case Study of Tongzhou, China

Abstract

The global epidemic is relatively stable, but urban pandemics will still exist. This study used sDNA (spatial design network analysis), spatial autocorrelation, and GWR (geographically weighted regression analysis) to identify potentially risky roads, pandemic hazard areas, and various infrastructure hazard areas in the Tongzhou District for urban sustainability. The results show that urban roads at risk during an epidemic have high proximity and aggregation effects. These roads are mainly concentrated in the core area. The hazard identification areas are focused on the urban sub-center and Yizhuang New Town. This paper derives the actual hazard areas using the POI (points of interest) data of COVID-19 (coronavirus disease 2019) and compares the results with the hazard identification areas. It is found that the hazard identification areas do not show the actual hazard area completely. In this study, GWR analyses based on gridded data of infrastructure POI proximity are used to obtain the hazard areas of various infrastructure types and develop different control ranges and methods. This provides new perspectives for identifying priority areas for epidemic prevention, control, and sustainable urban development.

1. Introduction

Since the emergence of COVID-19 (coronavirus disease 2019), which was triggered by a novel coronavirus, the economic and social order of cities worldwide has been significantly affected. Although the impact of COVID-19 is weakening, urban epidemics remain frequent. The fast spread of the disease characterizes this serious public health event, the widest range of infections, and the difficulty of prevention and control [1]. The World Health Organization (WHO) has reported 14.9 million COVID-19-related deaths in 2020–2022. The World Health Organization has warned countries worldwide about “X” diseases. The term “Disease X” is often used in research, modeling, and public health discussions to explore medicine’s disease control and management strategies. “Disease X” denotes an unforeseen and unidentified outbreak of an infectious disease. It represents the notion that a significant global pandemic could be triggered by a “pathogen X”, which, though currently unknown, has the potential to infect humans. “Disease X” is detrimental to the sustainability of the city. For example, COVID-19 could be a type of “Disease X”. This concept highlights the importance of preparedness for new and emerging infectious threats [2]. The movement and concentration of urban populations can exacerbate the outbreak of epidemics. Well-developed urban road networks greatly facilitate the movement of regional populations. However, this further increases the risk of epidemic spread and outbreak [3]. Before epidemic outbreaks, urban spaces with higher accessibility and proximity are the centers of the city and the areas of greater vitality in urban development [4,5]. After epidemic outbreaks, as these more vibrant places cluster more pedestrians, this becomes a space with a high degree of urban epidemic aggregation [6]. The study identifies the risk of epidemics in urban areas where road space and various types of infrastructure space are clustered. Based on the identification results, the article optimizes measures to contain further deterioration caused by epidemics [7].

The scope of this study is the Tongzhou District, Beijing. The Tongzhou District is the urban sub-center of Beijing. The Tongzhou District is among those in COVID-19 that were less affected by the outbreak in the early stages and recovered more slowly in the later stages. Although urban vitality increased in the middle and late stages, the level of development was still lower than the recovery pattern before the epidemic [8]. Ensuring the sustainability of urban recovery remains a challenge.

Past spatial studies of urban epidemic risk have focused on comprehensive demographic, economic, and social analyses. In 2022, Nazia’s team used standard deviation ellipses, spatial autocorrelation, and Geo Detector to analyze the geographic distribution and spatial heterogeneity characteristics of COVID-19 patients and deaths [9]. In 2022, Li Weiwei’s team used probes to analyze the geographical distribution of COVID-19 in China and the factors influencing it [10]. In the same year, Ming Sun’s team used a comprehensive overlay of the road network and facility POI data to predict the epidemic risk space in a big city and the overlay of facility mixing degree and population data to predict the epidemic risk space in a small town [11,12]. However, previous studies needed to have relevant research on the spatial impact of urban infrastructure on epidemics. In 2023, Rollier’s team used spatial regression modeling and spatio-temporal analysis models to explore the relationship between population mobility and the spatial spread of the COVID-19 virus in Belgium. The results revealed that virus transmission is faster in areas with high population mobility. In future virus prevention, managers should focus on controlling areas with high mobility [13]. In 2024, Li Yue’s team used a spatial Durbin model to analyze urban traffic flows during COVID-19 in the Glasgow area. The results show that traffic has an important impact on the city’s response to the crisis, urban resource allocation, and urban planning. However, epidemics have a direct effect on changes in urban traffic flows [14]. Previous epidemiological risk studies have focused more on analyzing simple population dispersions, simple transport systems, or simple facility spaces. These studies have not sufficiently analyzed the deeper causes of crowd movement within cities. For example, Ren and Jiao’s use of roads identifies hazardous areas that can only be matched to corroborate with part of the locations and does not provide a complete picture of the full epidemiological hazard area [11,12]. The movement of people in cities is often closely related to urban infrastructure and traffic. The spread of epidemics is also related to these two factors. Previous studies have only examined a single indicator without integrating the two for a comprehensive analysis. As a result, the results do not truly reflect the impact of urban operational mechanisms on the spread of epidemics.

This research examines the impact of various city infrastructures on the risk of urban disease transmission. Consequently, the study analyzes the pattern of epidemic disease spread in relation to the spatial distribution of urban facilities, focusing on urban planning to safeguard public health and mitigate epidemic risks sustainably. Utilizing the sDNA (spatial design network analysis) model, spatial autocorrelation, geographically weighted regression, and other analytical methods, the research explores how infrastructure influences epidemic risks across different urban areas. The goal is to identify high-risk areas associated with different types of facilities and to develop more effective outbreak control strategies.

The contributions and innovations of this study are mainly in the two aspects of indicator selection and research framework. In terms of indicator selection, this study integrates infrastructure point data with traffic indicators and embeds them in the fishnet to form an indicator of infrastructure proximity. This makes the infrastructure data in the fishnet not only reflect the distribution of facilities, but also the accessibility of each regional infrastructure. In terms of the research framework, this study integrates sDNA, GWR, and other methods to form a more comprehensive research system to obtain results that are more in line with the analysis of the epidemic hazard of facilities in the Tongzhou District. Managers can take targeted, sustainable management measures based on the epidemic hazard areas obtained for each type of infrastructure.

2. Materials and Methods

2.1. Research Area

This paper selected Tongzhou District in Beijing as the research object. As the seat of the Beijing government, the district takes over the administrative district functions of the old center of Beijing. Tongzhou District is located in the southeastern part of Beijing. The range is from 116°31′ to 116°56′ east longitude and 39°35′ to 40°01′ north latitude. Tongzhou District is connected to Beijing’s old city and neighboring areas such as Langfang in terms of location, with a high level of mobility of people and a high risk of virus transmission. The district boasts large logistics plants, numerous commercial complexes, convenient facility systems, abundant tourism resources, and rapidly developing economies. Tongzhou District’s traffic facilities include a railway station—the trains run within Beijing—and no airport. Tongzhou District’s external and internal transportation is mainly by road and metro. Tongzhou District has a relatively new traffic network compared to Beijing’s older urban areas. It may have a different resilience to respond to public health emergencies than traditional urban areas. Along with the modernization of urban development, Tongzhou District has a high population density and a significant intensity of crowd dispersal, which facilitates the spread of epidemics. Due to the special positioning of the district as a sub-center of Beijing, the infrastructure of Tongzhou District bears a greater potential for epidemic hazards. Since COVID-19, there have been more reports of epidemic risk in the district, which is of high research significance.

Tongzhou District has 22 towns. The towns are mainly clustered in the northwestern part of the district, as shown in Figure 1A. According to the planning, the district will accelerate the formation of a pattern of “one city, one belt, one axis, four districts, three towns and multiple points”. The district has two core areas and two major rivers. They are the Beijing sub-center, the Yizhuang New Town, the Chaobai River, which flows through the eastern border, and the Wenyu River, which flows through the territory, as shown in Figure 1B. The study area is the whole area of Tongzhou, with a total area of 889.03 square kilometers. This study uses the spatial autocorrelation results of road proximity and topological depth to confirm the extent of the hazard identification area. Taking the COVID-19 actual medium–high hazard area as a reference, this paper identifies the actual hazard area and analyzes the correlation between various types of infrastructure infections. This provides reference ideas for controlling urban epidemics in the new era.

2.2. Research Materials

The research data consist of line data (LOI) and point data (POI). These data were obtained by collecting information through traditional field visits and Internet big data. This thesis uses the medium–high hazard areas of Tongzhou District during the COVID-19 epidemic period of 2019–2022 as the epidemic data. Since the COVID-19 epidemic hazard data of Tongzhou District appeared from May to December 2022, which is a post-epidemic period, all data in this paper were captured during this period. The data are categorized into Tongzhou District urban road data (LOI data), epidemic risk point location data (POI data) for the whole district, and urban infrastructure data (POI data) for the entire district.

2.2.1. Urban Road Data

This paper crawled the road centerline data of the Tongzhou District in OpenStreet Map (https://www.openstreetmap.org/#map=9/7.2960/38.4137, accessed on 31 December 2022) in September 2022. This study removes roads outside the Tongzhou District boundaries and roads closed during the COVID-19 control period and fixes the deviated road centerline data based on satellite maps in GIS. After the above steps were completed, the data were plotted according to the axial mapping principle, as shown in Figure 2A. There are more than 30,000 road data points in the district. Tongzhou District has a well-developed and complete road network. The core areas, such as the urban sub-center, have a high density of roads. These road data are prepared to enter sDNA for spatial design network analysis.

2.2.2. Point Location Data of Medium–High Hazard Areas

This research collected all the data of medium–high hazard areas of the district for the period of COVID-19 through Sina APP’s COVID-19 real-time dynamic tracking data (https://news.sina.cn/zt_d/yiqing0121, accessed on 31 December 2022). The period of data collection was 1 May–13 December 2022, which is the period of intense occurrence of COVID-19 risks in Tongzhou District. This paper follows the residential medium–high hazard areas by locating the neighborhoods and villages regarding the POI latitude and longitude of the neighborhood committee (village council). Facilities such as companies, restaurants, and leisure facilities were located with the actual POI latitude and longitude. This paper deleted the POI data of medium–high hazard areas that did not appear in the study area of Tongzhou District in GIS. After data cleaning and latitude/longitude matching, the study took each medium–high hazard information as a point location. It sequentially obtained the point location data of medium–high hazard areas in the region. In this paper, a total of 13,180 epidemic hazard point locations were obtained, as shown in Figure 2B.

2.2.3. Point Location Data of Urban Infrastructure

This paper collected various urban infrastructure location data from AMAP (https://ditu.amap.com/, accessed on 31 December 2022) by using PYTHON 3.9 in September 2022 with Tongzhou District as the boundary. This paper selected infrastructure that may have an impact on urban epidemics. Infrastructure POI data that did not appear within the study area of Tongzhou District were removed by GIS. After data cleaning, data refinement, and coordinate transformation, a total of 59,648 points of interest were obtained in this paper, as shown in Figure 2C. These data were ready to be entered into GIS for analysis. These data contain 12 infrastructure categories: shopping facilities; restaurant facilities; company enterprise facilities; residential facilities; hotel facilities; scenic facilities; educational and cultural facilities; public facilities; healthcare facilities; life, sports, and leisure facilities; government and group facilities; and traffic facilities.

Tongzhou District has a higher percentage of commercial services categories such as shopping facilities; life, sports, and leisure facilities; restaurant facilities; and company enterprise facilities. Their proportions are 24.42%, 21.71%, 15.84%, and 12.99%, respectively. The sum totals 74.96% of the total number of categories. This result indicates that the urban vitality value of the district is very high and it is prone to rapid and large-scale spread of public health epidemics.

2.3. Methods

Under policy control, epidemic spread closely correlates with population distribution, built environment, and the disposition of various other spatial factors, such as roads and infrastructure layout [15]. Firstly, the study uses the construction of sDNA model and spatial autocorrelation analysis model to delineate hazard identification areas [16]. Second, the project uses the fishnet model and GWR model to analyze the scope of the actual hazard areas and the scope of various types of infrastructure aggregation spaces in the city, with attention to sustainability in the planning and management of these spaces [17,18,19,20,21,22,23]. Finally, after identifying potential hazard areas for various facilities, this paper develops graded control means based on the results. This study adopts multi-source data for more objective analysis and more detailed analysis of facility categories, contributing to the development of sustainable urban health strategies, as shown in Figure 3.

2.3.1. sDNA

This study uses the axial modeling method to establish the topological relationships of urban road networks based on the basic theory of spatial syntax. The method of sDNA spatial design network analysis based on inherited social network analysis method and improved spatial syntax method is chosen for this research. sDNA is usually used to analyze spatial networks such as urban roads and pedestrian spaces. This thesis uses the proximity (NQPDA) and topological depth (MADnc) from the sDNA results to calculate the road space in Tongzhou City. The results reflect that higher proximity means higher attractiveness and infection risk level; smaller topological depth means higher accessibility [24]. This study uses roads with high values of proximity (NQPDA) and low values of topological depth (MADnc) to indicate roads with higher accessibility and higher risk of epidemic transmission.

The proximity (NQPDA) parameter in the sDNA model represents how conveniently a road can reach other roads within the search radius. Roads with high NQPDA have high accessibility. Due to the global scale of the search radius in this study, a road with high NQPDA means a road with high accessibility in Tongzhou District. The formula is

$$\mathrm{NQPDA}={\displaystyle \sum _{\mathsf{y}\in {\mathsf{R}}_{\mathsf{x}}}\frac{\mathsf{p}(\mathsf{y})}{\mathsf{d}(\mathsf{x},\mathsf{y})}}$$

In which $\mathrm{NQPDA}\left(\mathrm{x}\right)$ is the proximity metric, $\mathrm{p}\left(\mathrm{y}\right)$ is the weight of node y within the search radius R, and $\mathrm{d}(\mathrm{x},\mathrm{y})$ is the shortest angular distance between points x and y.

The topological depth (MADnc) parameter in the sDNA model represents the degree of remoteness of a road to reach other roads within the search radius. Roads with a high MADnc have a high degree of remoteness and are more difficult to reach. In this study, roads with high MADnc imply roads at the edge of the Tongzhou District and areas that are not conveniently accessible. The formula is

$$\mathrm{MADnc}={\displaystyle \frac{{\displaystyle {\sum }_{\mathsf{y}\in {\mathsf{R}}_{\mathsf{X}}}{\mathsf{d}}_{\mathsf{M}}(\mathsf{x},\mathsf{y})\mathsf{P}(\mathsf{y})}}{{\displaystyle {\sum }_{\mathsf{y}\in {\mathsf{R}}_{\mathsf{X}}}\mathsf{P}(\mathsf{y})}}}$$

In which $\mathrm{MADnc}\left(\mathrm{x}\right)$ is the topological depth metric, $\mathrm{P}\left(\mathrm{y}\right)$ is the weight of the node y within the search radius, and ${\mathrm{d}}_{\mathrm{M}}\left(\mathrm{x},\mathrm{y}\right)$ is the shortest angular system distance between the x, y points.

2.3.2. Spatial Autocorrelation Analysis

After analyzing the global proximity indicator (NQPDA) and global topological depth indicator (MADnc) of the sDNA model of the road network in Tongzhou District, Beijing, spatial autocorrelation analysis is carried out in this study. Firstly, this section needs to verify the significance of the global spatial autocorrelation analysis by Moran’s I, Z-value, and p-value. Secondly, this section derives the clustering distribution through local spatial autocorrelation analysis. Spatial autocorrelation indicates the degree of correlation of a geographic phenomenon or attribute value with the same phenomenon or attribute value in neighboring spatial regions in the study area. This analysis can be divided into global and local spatial autocorrelations [25]. Global autocorrelation reflects the relationship of variables in the overall space. Local autocorrelation demonstrates the relationship between a single variable and the surrounding spatial elements [26]. Global autocorrelation can reflect if the proximity and topological depth indicators of roads in Tongzhou District are geographically correlated. The local spatial autocorrelation can reflect the high–high aggregation area of the proximity indicator and the low–low aggregation area of the topological depth indicator of the roads in Tongzhou District, which can then be superimposed to analyze the epidemic hazard identification area of Tongzhou District.

Global Spatial Autocorrelation

Global spatial autocorrelation is a study of the state of the spatial distribution of a particular attribute value over the entire study area. Global autocorrelation is an indicator that describes the relationship of a single variable in the overall space. The results are represented by Moran’s I coefficient. The formula is

$$\mathsf{I}={\displaystyle \frac{\mathsf{n}}{{\mathsf{S}}_0}}{\displaystyle \frac{{\sum }_{\mathsf{i}=1}^{\mathsf{n}}{\sum }_{\mathsf{j}=1}^{\mathsf{n}}{\mathsf{w}}_{\mathsf{i},\mathsf{j}}{\mathsf{z}}_{\mathsf{i}}{\mathsf{z}}_{\mathsf{j}}}{{\sum }_{\mathsf{i}=1}^{\mathsf{n}}{\mathsf{z}}_{\mathsf{i}}^2}}$$

In which ${\mathrm{z}}_{\mathrm{i}}$ is the deviation of element i from $({\mathsf{x}}_{\mathsf{i}}-\overline{\mathsf{X}})$, ${\mathrm{w}}_{\mathrm{i},\mathrm{j}}$ is the spatial weight of elements i and j, n is the total number of elements, and ${\mathrm{S}}_0$ is the sum of all spatial weights.

$${\mathsf{S}}_0={\displaystyle \sum _{\mathsf{i}=1}^{\mathsf{n}}{\displaystyle \sum _{\mathsf{j}=1}^{\mathsf{n}}{\mathsf{w}}_{\mathsf{i},\mathsf{j}}}}$$

Global spatial autocorrelation is a comprehensive index used to assess the degree of spatial autocorrelation in spatial autocorrelation analysis, which is usually measured by Moran’s I value. The Moran’s value takes the range of [−1, 1]. When the value is close to 1, it indicates a positive spatial correlation, indicating that the elements are spatially clustered, and the higher the value is, the more pronounced the spatial correlation is. When the value tends to −1, it means that the space presents a negative correlation, indicating that the elements are spatially discrete. The smaller the value is, the greater the spatial difference is. When the value is 0, it suggests that the space presents randomness.

Local Spatial Autocorrelation Analysis

Local spatial autocorrelation quantifies the local aggregation characteristics of a variable based on a global analysis statement. This identifies the specific distribution of an attribute within the study area.

Local Moran’s I formula is

$${\mathsf{I}}_{\mathsf{i}}={\displaystyle \frac{{\mathsf{x}}_{\mathsf{i}}-\overline{\mathsf{X}}}{{\mathsf{S}}_{\mathsf{i}}^2}}{\displaystyle \sum _{\mathsf{j}=1,\mathsf{j}\ne \mathsf{i}}^{\mathsf{n}}{\mathsf{w}}_{\mathsf{i},\mathsf{j}}({\mathsf{x}}_{\mathsf{i},\mathsf{j}}-\overline{\mathsf{X}})}$$

In which ${\mathrm{I}}_{\mathrm{i}}$ denotes the local Moran index of feature region i, ${\mathrm{x}}_{\mathrm{i}}$ is the attribute of element i, $\overline{\mathsf{X}}$ is the mean value of the corresponding attribute, ${\mathrm{w}}_{\mathrm{i},\mathrm{j}}$ is the spatial weight between the two elements i and j, and

$${\mathsf{S}}_{\mathsf{i}}^2=\frac{{\sum }_{\mathsf{j}=1,\mathsf{j}\ne \mathsf{i}}^{\mathsf{n}}{({\mathsf{x}}_{\mathsf{j}}-\overline{\mathsf{X}})}^2}{\mathsf{n}-1}$$

where n is the total number of elements.

This paper uses local spatial autocorrelation to identify clusters and abnormal regions of spatial features. The results are analyzed to reveal five main types of clustering distributions. These include high–high clustering (H–H), low–low clustering (L–L), high–low clustering (H–L), low–high clustering (L–H), and random distribution. H–H and L–L clustering denote spatially clustered distributional features. H–L and L–H clustering denote spatially dispersed distributional features. The random distribution indicates no spatial correlation in the distribution of elements.

2.3.3. Fishnet Analysis

This article uses the Create Fishnet tool in ArcGIS 10.4 to create 1 km × 1 km grid system in Tongzhou District, Beijing, and encodes a total of 1012 grids in all grids using the WGS_1984_UTM_Zone_50N projected coordinate system. The fishnet tool allows for the standardization of data into a uniform grid so that data from different areas are comparable. This reduces the differences due to the size and shape of the study area. The tool can be adapted to the needs of the study scale. For this study, the 15 min living circle distance—the distance that can be reached on foot within 15 min—was chosen as a grid scale. Fishnet analysis can be easily combined with other spatial analysis methods for adequate spatial analysis, such as spatial autocorrelation, GLS, and GWR. This paper calculates the number of medium–high risk points in each grid and the sum of the proximity of each facility separately for further analysis [27,28].

2.3.4. Geographically Weighted Regression Analysis (GWR)

The GWR model is a local spatial analysis method [29]. This model is used in studies such as geography and urban spatial analysis. The GWR model integrates linear regression models with spatial location elements to construct local regression equations in order to obtain results that better reflect the actual situation. As the spatial distribution of epidemics varies across facilities, this paper performs geographically weighted regression (GWR) analyses by linking the epidemic hazard data from the fishnet analyses with the facility proximity data. This allows for an analysis of the extent to which different facilities in the study area affect different locations within the potential area of the epidemic (GWR regression coefficients). This allows for analyzing the scale of impact of different facilities in the study area on different locations within the epidemic potential area. The higher the regression coefficients of the facilities in the grid, the greater the risk space of the facilities. This article is based on the geographically weighted regression analysis tool in ArcGIS, choosing AICc for bandwidth and adaptive methods for kernel type. After GWR analysis, global spatial autocorrelation analysis is performed on the residuals to determine the potential risk space of each facility after observing the correct model fit state [30,31,32,33]. The formula for GWR is

$${\mathsf{Y}}_{\mathsf{i}}={\mathsf{\beta }}_0({\mathsf{u}}_{\mathsf{i}},{\mathsf{v}}_{\mathsf{i}})+{\displaystyle \sum _{\mathsf{k}=1}^{\mathsf{p}}{\mathsf{\beta }}_{\mathsf{k}}({\mathsf{u}}_{\mathsf{i}},{\mathsf{v}}_{\mathsf{i}})}{\mathsf{X}}_{\mathsf{i}\mathsf{k}}+{\mathsf{\epsilon }}_{\mathsf{i}}$$

In which ${\mathrm{Y}}_{\mathrm{i}}$ is the value of the dependent variable attribute of sample i. In this study, it is the number of points of risk, and ${\mathsf{\beta }}_0{(\mathrm{u}}_{\mathrm{i}}{,\mathrm{v}}_{\mathrm{i}})$ is a constant term; ${\mathsf{\beta }}_{\mathrm{k}}{(\mathrm{u}}_{\mathrm{i}}{,\mathrm{v}}_{\mathrm{i}})$ is the regression coefficient of the the kth explanatory variable of sample I; ${\mathrm{X}}_{\mathrm{ik}}$ is the value of the kth explanatory variable of sample i, and ${\mathsf{\epsilon }}_{\mathrm{i}}$ is the random error.

3. Results

3.1. Analysis of Hazard Identification Areas

3.1.1. Result of sDNA

This section uses the axial method to model the road spatial design network during COVID-19 in the Tongzhou District, Beijing. It is used to analyze the spatial situation of urban roads under the state of local road control or closure during the epidemic. In this paper, the constructed road model is analyzed by sDNA. The parameters use a hybrid metric combining angular and metric distances, and an infinite distance (N) is chosen as the search radius. The analysis results are shown in Figure 4A for global proximity (NQPDA) and in Figure 4B for global topological depth (MADnc).

The results show that roads in the core area of the Tongzhou District (Beijing sub-center area) show more obvious aggregation. The Beijing sub-center area and Yizhuang New Town (Tongzhou area) show a more substantial aggregation effect and are clustered along major roads. Aggregation space is more susceptible to infection risk spaces. The clustered roads provide conditions for the rapid and large-scale spread of public health epidemics.

Global proximity reflects the attractiveness of the urban road network, with larger values indicating more significant potential for attracting people and traffic and greater aggregation. In areas with high global proximity, roads are colored in red. Conversely, roads are colored in blue. The highest value of global proximity is 50.67, and the lowest is 11.14. The global proximity of the Tongzhou District is overall high, as shown in Figure 4A. The whole district has the potential for aggregation and infection. The urban sub-center and Yizhuang New Town (Tongzhou part) areas have the highest global proximity of roads, which is more likely to create a risk of aggregation and infection. Therefore, crowd and traffic monitoring should focus on high-road proximity areas. Global topological depth indicates the accessibility of the urban road network, with smaller values indicating higher accessibility and more convenient traffic.

Roads in areas with low global topological depth are represented in blue, and conversely, roads are in red. Red roads mean less accessibility and less risk of epidemic spread. In this study, the highest value of global topological depth is 2481.31, and the lowest value is 475.74, which is analyzed by sDNA calculation. The road network accessibility is higher in the Tongzhou District as a whole. The urban sub-center has low topological depth and the highest accessibility; Yizhuang New Twon is the second, and the edge of the study area has high topological depth. The difficulty of accessibility is more significant, as shown in Figure 4B. Through the mathematical visualization model, global topological depth is negatively correlated with global proximity. This indicates that the lower the global topological depth of the road, the higher the global proximity, the higher the accessibility of the road, and the stronger the attraction and aggregation effects. Both the R² result and the adjusted R² result are 0.91. The two sets of data are highly negatively correlated, as shown in Figure 5. This implies that roads with low topological depth and roads with high proximity are more compatible in the Tongzhou District. These areas have great potential to spread epidemics and, therefore, require sustainable traffic management measures to reduce the spread risk.

The above analysis shows that the roads in cities with the highest risk of epidemic transmission and high infection potential are usually located in the core areas of urban development. Roads with a low risk of infection are usually located in the peripheral areas with a low level of aggregation.

3.1.2. Spatial Autocorrelation Analysis Results

Global Spatial Autocorrelation Analysis

The global spatial autocorrelation of proximity and topological depth is analyzed as shown in

Table 1. Global spatial autocorrelation analysis attribute values of sDNA model.

Analytical Projects	Global Moran’s I	Global Z(I)	Global p-Value	Spatial Clustering
NQPDA	0.53	420.85	0.00	YES
MADnc	0.51	411.69	0.00	YES

. The global spatial autocorrelation Moran’s I of road network proximity and topological depth during COVID-19 in the Tongzhou District is 0.53 and 0.51 > 0. The global Z(I) scores are 420.85 and 411.69, respectively, much larger than 2.58, with global p-values of 0.00. This indicates that the road network spatial exhibits a clustering trend, and the clustering effect is noticeable. This trend forms a prominent aggregation space, which shows a significant positive correlation with the aggregation of urban spatial structures. Therefore, there are apparent epidemic risk spaces in the district during COVID-19, and appropriate planning measures should be taken to manage the potential risk spaces to achieve adequate safety prevention and control.

Local Spatial Autocorrelation Analysis

The global spatial autocorrelation table reveals that the epidemic risk area in a particular region shows clustering, and the local spatial autocorrelation analysis further identifies the specific location of the clustered area [34]. Clustering and outlier analyses are performed on the basis of proximity and topological depth global spatial autocorrelation analyses, and local spatial autocorrelation analysis results are obtained in this study.

The results are sequentially displayed in the local spatial autocorrelation analysis map, as shown in Figure 6. The results show five main clusters: high–high, low–low, high–low, low–high, and insignificant. The proximity results form significant high–high clusters in the urban sub-center and Yizhuang New Town. This indicates that the region has higher accessibility and is more likely to form clustered spatial. Meanwhile, the topological depth results form more prominent low–low clusters in the urban sub-center and Yizhuang New Town. This indicates that the region has a shorter topological distance, is more accessible, and has a more tremendous potential for infection. The spatial autocorrelation of proximity and topological depth in the fringe area of the Tongzhou District is low–low clustering and high–high clustering, respectively. This indicates that roads in this region are more difficult to reach. The two spatial autocorrelation results for the south–central part of the Tongzhou District are high–low clustering, low–high clustering, and insignificant areas. These regions are in the excessive stage of high–high clustering and low–low clustering.

3.1.3. Delineation of Hazard Identification Area Based on Spatial Autocorrelation Analysis

Through the above analyses, this study divides the hazard identification area into three categories: hazard area, buffer area, and low-hazard area, as shown in Figure 7. The spatial autocorrelation is used in this article to classify the hazard identification area into three types of areas, which are hazard area, buffer area, and low-hazard area. The proximity of a local spatial autocorrelation high–high clustering region and a topological depth local autocorrelation low–low clustering region forms the hazard area. The proximity of a local spatial autocorrelation low–low clustering region and the topological depth local autocorrelation high–high clustering region form the buffer area. The low-hazard area is formed by areas of high–low clustering and low–high clustering spatially formed by other insignificant areas.

The hazard areas are mainly concentrated in the urban sub-center of the Tongzhou District and the Yizhuang New Town area, as shown in the red area in Figure 7. A high population density, abundant commercial centers, and diverse industrial functions characterize the hazard area. It is mainly located in the core area of the Tongzhou District. Low-hazard areas are primarily concentrated in the edge areas of the Tongzhou District, as shown in the blue areas in Figure 7. These areas are characterized by weak accessibility and relatively low intensity of crowd dispersion. They are more discrete in their distribution, unlike the hazard area, which is spread out in a contiguous pattern. The hazard buffer area is located between the hazard area and low-hazard area, as shown in the yellow area of Figure 7. This area has a medium-risk clustering effect with some degree of epidemic risk. The buffer area connects the high-risk core area and the peripheral low-risk area and serves as a transition. This area does not have significant spatial correlation.

3.2. Analysis of the Actual Hazard Area

The above analyses reveal that the road space in the Tongzhou District is clustered and reflects the district’s epidemic hazard identification area [35]. In this section, the fishnet tool is used to analyze the actual hazard area of the epidemic within the study area. Firstly, this study divides the Tongzhou District into 1000 m × 1000 m grids on average and confirms the extent of the actual hazard area after connecting the medium–high hazard points of COVID-19. Secondly, this paper compares the scope of the actual hazard area with the hazard identification area to confirm the degree of coincidence between the two.

3.2.1. Delineation of Actual Hazard Area Based on Fishnet Analysis

This section constructs grids based on the POI data of the epidemic risk points for the fishnet analysis. According to the number of epidemic points in each grid, this section can determine the scope of the actual hazard area. There are a total of 320 grids with at least one medium–high risk point from 1 May–13 December 2022 within the scope of this study, representing 31.62% of the total number of grids. Firstly, this paper adopts the natural break method to divide the grid map of risk point locations into two types of actual hazard areas. This chapter takes the value 91–459 area with the strongest clustering effect, as a high hazard area, and takes the value 1–90 area with a stronger clustering effect, as a medium hazard area, as shown in Figure 8A. The actual hazard area is obviously concentrated in the northwestern part of the study area close to the downtown area of Beijing. The area is distributed into the urban sub-center on both sides of the Wenyu River and is mainly concentrated in 20 towns, such as Xinhua, Beiyuan, Yangzhuang, and Jiukeshu in the urban sub-center and Yongshun, Yizhuang New Town, Songzhuang, and Majuqiao. Secondly, this study subjected the actual hazard area data to a local spatial autocorrelation analysis. The results show that the high–high clustering range of the actual hazard area is basically located in the core area of the whole region, near commercial streets, enterprise campuses, and scenic areas, as shown in Figure 8B. The distribution of the hazard area is not related to natural topography such as rivers. According to the results of the fishnet analysis, the total area of the actual hazard area in the Tongzhou District is 32,000 hectares.

3.2.2. Comparison of Actual Hazard Area and Hazard Identification Area

This section uses the clip function of ArcGIS to clip the intersection area between the hazard identification area and the actual hazard area analyzed by local spatial autocorrelation. The results of the statistical calculation are shown in

Table 2. Comparison of hazard identification area and actual hazard area.

Analytical Projects	Hazard Identification Area	Actual Hazard Area
Area (ha)	21,327	32,000
Intersection area (ha)	11,803	11,803
Percentage of intersection area (%)	55.34	36.88

. This paper concludes that the hazard identification area is 21,327 hectares and the actual hazard area is 32,000 hectares. The actual hazard area is larger than the hazard identification area. The intersection area of the two is 11,803 hectares, accounting for 55.34% of the hazard identification area and 36.88% of the actual hazard area. This indicates that the hazard identification area delineated by the results of the spatial autocorrelation analysis of road proximity and topological depth is limited and cannot fully show the actual hazard area. The analysis of epidemiological space by road spatial elements alone is unilateral. This study should introduce the integration of infrastructure spatial layout data with road data. This will improve the accuracy and sustainability of urban spatial epidemiological hazard management.

The distribution of the intersection of the two with the hazard identification area and the actual hazard area is shown in Figure 9. Since Wenyu River and its two banks have high population density, high economic vitality and strong aggregation of cities, the hazard area in this region is accurately identified. The hazard areas of Yizhuang and Songzhuang, which are far from the river, need to be clearly identified. The reason for this phenomenon may be due to a direct relationship between crowd aggregation and dispersal activities and road proximity in the areas on both sides of the Wenyu River. This leads to accurate epidemiological hazard identification. Road proximity has less influence on the spread of epidemics in areas farther from the river, such as Yizhuang and Songzhuang. Other elements, such as facilities, have a greater impact on such areas, thus leading to inaccurate identification.

3.3. Identification and Control of Epidemic Hazards in Urban Infrastructure

Since the hazard identification area derived from spatial autocorrelation does not match the actual hazard area, this paper needs to identify the hidden hazards of the clustering range of the proximity of each type of infrastructure with the actual hazard area and identify the hazard area of the whole domain for each type of infrastructure. Based on the identification results, we confirmed sustainability measures to reduce the risk of a long-term epidemic.

3.3.1. Fishnet Analysis of Infrastructure Proximity Aggregation Range

This study superimposes the quantity of facilities and road proximity metrics and then connects them in grids as a representation of infrastructure proximity. The infrastructure proximity variables are divided into 12 categories, which are proximity to facilities such as shopping, restaurants, companies, and enterprises. This section counts the infrastructure proximity grid quantity, the sum value of each type of infrastructure proximity, and the average proximity value for each type of infrastructure. The results are as shown in

Table 3. Infrastructure proximity results.

Category	Grid Quantity	Sum Value	Average Value
Company Enterprise	704	267,938.54	380.59
Life, Sports, and Leisure	686	464,568.80	677.21
Shopping	614	519,157.34	845.53
Government and Group	561	146,099.53	260.43
Public Facility	473	49,103.21	103.81
Education and Culture	435	144,484.58	332.15
Restaurant	422	335,337.42	794.64
Healthcare	364	78,021.71	214.35
Residential Facility	335	66,977.76	199.93
Scenic Facility	237	14,290.93	60.30
Hotel	167	24,183.99	144.81
Traffic Facility	47	3386.38	72.05

.

The infrastructure proximity grid quantity can be interpreted as the infrastructure distribution. Grids with facility proximity values > 0 are those with the presence of that type of facility. Facilities with a high number of grids are widely distributed. Facilities with a low number of grids are distributed centrally. In the distribution of infrastructure proximity, eight types of facilities, including restaurants, company enterprises, and educational and cultural facilities, are widely distributed in the whole area of the Tongzhou District. Among them, company enterprise facilities; life, sports, and leisure facilities; and shopping facilities are the most widely distributed. Their grid numbers are 704, 686, and 614, respectively, and these three categories account for more than 60% of the total number of grids. Scenic, hotel, residential, and traffic facilities are more closely distributed. The number of grids for traffic facilities is 47, and the number of grids for these facilities accounts for 4.64% of the total number of grids.

The sum value of each infrastructure proximity can be interpreted as a combination of the number and proximity of each type of facility. The commercial types of infrastructure have the highest quantity and the highest proximity values. The sum of proximity is high for shopping facilities; life, sports, and leisure facilities; restaurant facilities; and company business facilities. Their values are much larger than the other facilities. The high numbers and proximity values of commercial-type facilities in the Tongzhou District has a high epidemiological risk.

The average proximity value for each type of infrastructure can be interpreted as the location of facility distribution. Facilities with high values are distributed in urban core areas with high road proximity such as urban sub-center and Yizhuang New Town. These areas have higher epidemic risks. The results indicate that various types of infrastructure are mainly clustered in the core area of the Tongzhou District. The average proximity of commercial-type facilities and public service-type facilities is higher. These facilities are more distributed in the urban core area.

3.3.2. Identification and Control of Infrastructure Epidemic Hazards within Tongzhou District

This research uses the ArcGIS clip function to cut out the intersection area between the proximity aggregation range and the actual hazard area of each type of infrastructure. It calculates the proportion of the intersection area to each type of facility proximity aggregation range and the actual hazard area respectively, as shown in

Table 4. Percentage of intersection area with various types of infrastructure proximity clustering area and actual hazard area.

Category	Amount	Percentage (%)
Shopping	14,569	24.42
Restaurant	9451	15.84
Company Enterprise	7748	12.99
Residential Facility	1891	3.17
Hotel	676	1.13
Scenic Facility	424	0.71
Education and Culture	4108	6.89
Public Facility	1450	2.43
Healthcare	2155	3.61
Life, Sports, and Leisure	12,951	21.71
Government and Group	4127	6.92
Traffic Facility	98	0.16

. After scatter fitting the two data groups, this study reveals that the two are negatively correlated. The results indicate that the higher the percentage of intersecting areas to the actual hazard area, the lower the percentage of intersecting areas to this type of infrastructure area. This reflects that the more dispersed the distribution of this type of infrastructure is within the city, the more dispersed it is within the actual hazard area.

For example, restaurant facilities; shopping facilities; company enterprise facilities; educational and cultural facilities; public facilities; healthcare facilities; life, sport, and leisure facilities; government and group facilities; and residential facilities, are relatively dispersed. For such infrastructure, management needs to focus more on monitoring the facilities within the actual hazard area. It does not need to monitor the facilities outside the actual hazard area too much. On the contrary, this study found that the higher the percentage of intersecting areas in the infrastructure area of this type, the lower the rate of the actual hazard area, and the distribution is relatively concentrated in the actual hazard area. For example, hotel facilities and scenic facilities are relatively concentrated. Since these facilities are in a higher percentage of the hazard area, management should strengthen the ability of region-wide coordination and control for this type of high-risk infrastructure and monitor it by region-wide control. Transportation facilities have less data in both groups and lower infection potential, as management can reduce the allocation of monitoring and control resources for this type of infrastructure. The above analyses reflect the mechanism of infrastructure epidemic analysis for the whole area of the Tongzhou District. The decentralized distribution of infrastructure results in a low percentage of intersections within the actual hazard area, while the centralized distribution of infrastructure results in a high percentage of intersections within the actual hazard area. Therefore, this distribution characteristic needs to be taken into account in management and surveillance strategies.

3.3.3. Identification and Control of Epidemic Hazards for Infrastructure in the Actual Hazard Area

GWR Model Indicators Analysis

This section uses the proximity data for the 12 types of infrastructure data within the actual hazard area as the independent variable and the number of medium–high hazards within each grid as the dependent variable for the OLS and geographically weighted regression analyses [36,37]. As shown in

Table 5. GWR model and OLS model to analyze indicator results.

Indicator Name	GWR	OLS
Adjusted R2	0.62	0.56
AICc	3252.26	3293.83
Bandwidths	181	\
Effective number of parameters	47.06	\
Sum of squares of the residuals	372,471.96	\

, the adjusted R² of the GWR model is 0.62. The adjusted R² of the OLS model is 0.62. The GWR model has a stronger and better fit than the OLS model. The bandwidth parameter of the GWR model is chosen as AICc. AICc can balance the goodness-of-fit and the number of parameters during the analysis. This can find the optimal bandwidth of action to avoid overfitting the results. The AICc of the GWR model is 3252.26, which is smaller than the OLS model. This indicates that the GWR model is stronger than the OLS model in terms of performance. The GWR model is more realistic and better in the analysis of the study in this section. The GWR model kernel selects adaptive bandwidth analysis. This model improves the flexibility and accuracy of the model, better reflects the spatial heterogeneity of the results, and improves the interpretation of the regression results. The bandwidth of action in the analysis of the GWR model is 181, which is 56.56% of the total sample size, which is about half the global scale. This indicates spatial heterogeneity in the impact of every kind of infrastructure on the epidemic.

This article analyzes the spatial autocorrelation of the residuals obtained after the GWR analysis. The results conclude that the residuals of the model analysis results are random. This indicates that the results of this model are correct and well fitted.

The results of the regression coefficients of the GWR model are shown in

Table 6. GWR model regression coefficient results.

Categories	Average Value	Standard Deviation	Minimum Value	Maximum Values
Intercept	18.72	11.04	7.16	42.04
Shopping	0.01	0.01	−0.01	0.01
Restaurant	0.01	0.01	0.00	0.02
Company Enterprise	−0.01	0.01	−0.03	0.00
Residential Facility	0.14	0.06	0.04	0.23
Hotel	0.02	0.03	−0.04	0.08
Scenic Facility	−0.09	0.05	−0.23	−0.02
Education and Culture	−0.01	0.02	−0.03	0.05
Public Facility	−0.02	0.02	−0.06	0.01
Healthcare	0.01	0.01	−0.02	0.03
Life, Sports, and Leisure	−0.01	0.01	−0.03	0.02
Government and Group	0.01	0.01	−0.01	0.04
Traffic Facility	0.07	0.06	−0.04	0.22

. The results show that intercept is the factor that has the greatest positive effect on the epidemic. Intercept can be interpreted as an effect of locational factors. The regression coefficient of this variable is 18.72, which is much larger than the effect of other variables. This indicates that location has a significant positive effect on epidemics in the Tongzhou District. The average value of the coefficient of proximity to seven types of infrastructure with positive results are shopping, restaurant, residential, hotel, healthcare, governmental and group, and traffic facilities. These seven categories of facilities have a positive impact effect on the epidemic. The facilities with a high impact are residential facilities, traffic facilities, and hotel facilities. The average values of the regression coefficients were 0.14, 0.07, and 0.02, respectively. The mean results of the coefficients of the proximity of five types of infrastructure are negative, namely, company enterprise facilities, scenic facilities, educational and cultural facilities, public facilities, and life, sports, and leisure facilities. Due to urban control and other reasons, these facilities are less affected by epidemics due to closure measures during epidemics. Therefore, these infrastructures are negatively associated with epidemic risk. The facilities with the greatest negative impact were scenic facilities. The average value of the regression coefficient is −0.09.

Spatial Analysis of GWR Model

Based on the distribution of regression coefficients for infrastructure positively influencing epidemics in the actual hazard areas of the Tongzhou District, this section identifies the hazard areas for each type of facility and develops management strategies [38,39,40]. The results indicate that locations with high facility regression coefficients are strongly and positively associated with epidemic risk. These locations are potential hazard areas for facilities. In this article, the regression coefficients are classified into five categories based on the natural break method [41].

The two categories of grids with positive correlation coefficients between the facility and the actual hazard area are indicated by red and orange colors. They are the high-hazard areas of the facility. The three categories of grids with negative or insignificant coefficients between facilities and medium–high potential hazards are represented by blue, green, and yellow colors, respectively. They are the lower-hazard areas of the facility. Towns with a high epidemic risk for each type of infrastructure are indicated by red dots in the figure. The variables are discussed in this section from strongest to weakest, according to the degree to which they affect the epidemic.

(1) Removing the effect of proximity to various types of infrastructure, this section obtains the intercept variable that has the largest positive effect on the epidemic. These data can be interpreted as the effect of location on epidemics. The regression coefficients of intercept are shown in Figure 10A: the regression coefficients of this variable are 7.16–33.37. This indicates that all locations within the actual hazard area are positively correlated with the risk of epidemics. The largest regression coefficients were found for towns in the urban sub-center area and Songzhuang Town. These locations are strongly positively correlated with epidemic risk. These are hazard areas that are more likely to have epidemics. Therefore, in terms of epidemic prevention and control, management should increase the hazard monitoring of the infrastructures in these locations.

(2) The distribution of regression coefficients for residential facilities is shown in Figure 10B: the regression coefficients for this variable range from 0.04 to 0.23. This indicates that all residential facilities in the actual hazardous area are positively correlated with the risk of epidemics. The regression coefficients are higher for residential facilities in 18 towns in the central and northern parts of the Tongzhou District. Residential facilities in these towns have a high epidemic hazard and are clustered in the urban sub-center area. Residential facilities in these locations are strongly and positively associated with epidemic hazard. These are the hazard areas for residential facilities.

(3) The distribution of regression coefficients for traffic facilities as shown in Figure 11A: the regression coefficients for this variable are −0.04–0.22. This reveals that some of the traffic facilities in the actual hazard area are positively correlated with the risk of epidemics. The regression coefficients for traffic facilities are higher in 17 towns in the central part of the Tongzhou District. Traffic facilities in these towns are strongly and positively associated with epidemic hazard. These are hazard areas for transport facilities.

(4) The distribution of regression coefficients for hotel facilities as shown in Figure 11B: the regression coefficients for this variable are −0.04–0.08. This indicates that some hotel facilities in the actual hazard area are positively correlated with epidemic risk. The regression coefficients for hotel facilities are higher in 13 towns in the western part of the Tongzhou District. Hotel facilities in these towns are strongly and positively associated with epidemic hazard. These are the hazard areas for hotel facilities.

As shown in Figure 10 and Figure 11, the high-hazard areas for the intercept and these three types of facilities are different. The highest hazard areas are located in the northern part of the Tongzhou District. The highest hazard areas for the three types of facilities with the greatest positive epidemiological impact are clustered in the urban core in the western part of the Tongzhou District. The hazard areas for residential facilities are concentrated in the northwestern part of the Tongzhou District and on both sides of the Wenyu River sub-center section. The hazard areas for traffic facilities are mainly clustered in the central part of the Tongzhou District and distributed along both sides of the Wenyu River sub-center section of the city. The hazard areas for hotel facilities are located on the west side of the Tongzhou District and are distributed on the southwest side of the Wenyu River.

(5) The distribution of regression coefficients for government and group facilities is shown in Figure 12A: the regression coefficients for this variable are −0.01–0.04. This means that some of the government and group facilities in the actual hazard area are positively associated with epidemic risk. The regression coefficients for government and group facilities are higher in the 10 towns in the south and north of the Tongzhou District. They are distributed far from the core area of the Tongzhou District. Government and group facilities in these towns are strongly and positively associated with epidemic risk. These are the hazard areas of government and group facilities.

(6) The distribution of regression coefficients for healthcare facilities is shown in Figure 12B: the regression coefficients for this variable are −0.02–0.02. This indicates that some of the healthcare facilities in the actual hazard area are positively correlated with the risk of epidemics. The regression coefficients are higher for healthcare facilities in 13 towns in the northeastern part of the Tongzhou District. The facilities of healthcare facilities in these towns are strongly and positively associated with the epidemic hazard. These are the hazard areas of healthcare facilities.

(7) The distribution of regression coefficients for shopping facilities, as shown in Figure 13A: the regression coefficients for this variable are −0.01–0.02. This shows that some of the shopping facilities in the actual hazard area are positively correlated with the risk of epidemics. The regression coefficients for shopping facilities are higher in 18 towns in the northwestern part of the Tongzhou District. These towns are mainly located in the urban sub-center area, which is strongly and positively associated with epidemic hazard. These are the hazard areas for shopping facilities.

(8) The distribution of the regression coefficients of restaurant facilities is shown in Figure 13B: the regression coefficients of this variable are 0.00–0.02. This indicates that all the restaurant facilities in the actual hazard area are positively correlated with epidemic risk. The regression coefficients for restaurant facilities are higher in the 14 towns in the northeastern part of the Tongzhou District. Restaurant facilities in these towns are strongly and positively associated with the epidemic hazard. These are the hazard areas of restaurant facilities.

As shown in Figure 12 and Figure 13, the hazard areas of the four facilities that have a large positive impact on the epidemic are not uniform. The hazard areas for government and group facilities are primarily clustered on the south side of the Tongzhou District and distributed along the southwest side of the Wenyu River. The hazard areas for healthcare facilities are mainly clustered in the northeast side of the Tongzhou District and distributed along the northeast side of the Wenyu River. The shopping facilities hazard areas are primarily clustered on the northwest side of the Tongzhou District, and the clustering is less related to the distribution of the river. The hazard areas for restaurant facilities are primarily clustered on the east and north sides of the Tongzhou District and are located primarily in the northeast of the Wenyu River.

In summary, the spatial distribution of risk for various types of infrastructure is closely related to administrative and functional zoning and natural conditions such as rivers. At the same time, the infrastructure hazard area also has a view on the distribution form of facility proximity. The Tongzhou District government can coordinate and control all kinds of facility hazard areas based on functional zoning or natural conditions such as rivers. Each town government based on the town area identified the actual hazard area of various types of infrastructure return coefficient for higher facility hazard area, to carry out targeted control. This will promote the sustainability of urban health in the Tongzhou District area.

4. Discussion

This paper uses physical spatial data such as roads and infrastructure and indicators such as road proximity, topological depth, and spatial autocorrelation. This research uses fishnet and geographically weighted regression to analyze the actual hazard areas, ensuring sustainable urban health planning. This thesis takes the Tongzhou District in Beijing as the study area, and the following discussion is derived from the results of the above analyses:

Epidemic hazard area analysis: This paper analyzes the road network in Tongzhou District using sDNA and spatial autocorrelation to derive epidemic hazard identification areas prone to congestion and with high epidemiological potential. It was found that areas with high road accessibility and low topological depth, especially the urban sub-center and Yizhuang New Town, are high hazard areas. In contrast, low hazard areas are located in marginal areas with low population activity. This study analyzes the COVID-19 hazard data using the fishnet tool to derive the actual hazard zones. Due to the relaxation of control measures, there were increased medium- and high-risk points at the end of 2022, and the high hazard areas were concentrated in the urban sub-center, Yizhuang and Songzhuang. Comparing the hazard identification area obtained by spatial autocorrelation analysis with the actual hazard area derived from fishnet analysis, it is clear that a spatial analysis of roads alone is not sufficient [11,12], especially in areas such as Songzhuang.

Infrastructure epidemic hazard analysis: Factors such as roads, hospitals, schools, and shopping centers are associated with epidemic incidence [42]. This suggests that infrastructure factors may be associated with epidemic transmission. In this paper, a preliminary infrastructure hazard control program for the entire region is developed by analyzing the area where the range of proximity aggregation of each type of infrastructure intersects the actual hazardous area with the scatter fit analysis of the proportion of each of the two. This study uses the GWR model for infrastructure risk identification in the actual hazard area. Eight variables, including residential and traffic facilities, are found to be positively associated with epidemic hazard (the regression coefficients are positive). This may be related to the pattern of activities during the epidemic and the management style of the government. Finally, the spatial analyses of these variables in this study revealed different hazard areas, some associated with rivers and administrative areas and others with functional areas. This study identifies more precisely the location of epidemiological risk for each type of facility. District and town governments can develop different control strategies for the results of hazard identification for each type of infrastructure in the actual hazard area. This will enhance the sustainability of urban resilience and healthy development.

This study has implications for the management of epidemics and sustainable public health development in other cities. Other cities can strengthen epidemic prevention and management by targeting areas with high road accessibility and high infrastructure aggregation such as restaurants and shopping. In urban planning, managers can increase the number of medical facilities and epidemiological sites in these areas. This would ensure that in the event of an epidemic outbreak, healthcare facilities can be used quickly to slow down the spread of the epidemic. The combination of geographically weighted regression and fishnet in this study can identify high hazard areas for various types of infrastructure. Other cities can use this approach to analyze multiple types of spatial data to develop epidemic management measures that have less impact on the functioning of the city.

However, this study still has limitations. Firstly, the epidemiological data in the study only partially reflect the transmission process due to the different ways of classifying medium- and high-risk areas at each stage, and it is not possible to distinguish whether the epidemiological data are externally imported or internally transmitted. Future studies need more accurate access to data. Second, the epidemic in the Tongzhou District from the end of 2019 to the middle of 2022 was strictly managed, with less crowd activity and lower epidemic transmission. Epidemic data—COVID-19 hazard area data—appeared centrally in the second half of 2022, and the study results can only reflect the epidemiological potential due to crowd dispersal and facility activity in the Tongzhou District in the post-epidemic period. The study’s results do not provide a comprehensive analysis of infection in the early stages of the epidemic. Although the results of the analysis do not accurately reflect the early epidemic risk, managers can use the results of this thesis to predict the scope of the early epidemic potential and risk prevention. Third, the infrastructure POI data crawl is a September 2022 cross-section. This only takes into account the phase of post-epidemic COVID-19 when it was most infectious and does not account for differences between phases. Future research could refine the data to consider the effects of spatio-temporal variation. Research should analyze differences in the risk of epidemic transmission by infrastructure over time and under different management models. Fourth, in terms of data, future studies could add data such as the intensity of crowd dispersion and nighttime lighting to enrich the number of variables analyzed. In terms of model analysis, future studies can introduce the multi-scale geographically weighted regression model (MGWR) and SHAP model [43] for analysis and compare the effects with the results of the GWR model and OLS model. Finally, some of the analyzed results are not consistent with conventional understanding. For example, infrastructure variables such as life, sports, and leisure facilities are negatively correlated with epidemic hazard in the model, which is inconsistent with common sense. Future studies should explore these anomalous results and their impact mechanisms in detail.

5. Conclusions

This research identifies epidemic hazards through urban roads and infrastructures. And this study compares actual hazard areas with hazard identification areas to assess the effectiveness of using road space alone for hazard identification. The results show that the identification of epidemic risk by using road space alone is inaccurate. This study proposes a sustainable method for identifying and controlling epidemiological risks in the whole area of the Tongzhou District based on infrastructure proximity. It uses the GWR model to identify and control the infrastructure within the actual hazard area.

This study expands the study of epidemic hazards in different types of urban infrastructure, provides new perspectives for future sustainability epidemic prevention and control, and provides a new method for analysis of epidemic risk in other cities. By combining the spatial design network model and the GWR model, it breaks through the limitation of identifying only hazard areas. Compared with initial studies that relied only on data such as POIs and roads to identify aggregations to predict epidemiological hazard areas and simple geostatistical analyses using actual endemic data [44], this study achieves refined, hierarchical control, and optimization of individual facilities.

This study reveals the impact of infrastructure layout and facility proximity on epidemic transmission through a multi-method and multi-data fusion analysis. Related research such as urban planning and public health can be further validated in more diverse areas by combining accurate data to enhance the accuracy and usefulness of the research model. Based on the results of this study, city managers can make rational healthcare facility layouts and pedestrian and road management in urban planning for areas at high risk of epidemics. City managers can focus on prevention in areas of the facilities that are at higher risk for epidemics. This can enhance the resilience of cities against epidemics and contribute to long-term urban epidemic management.