Spatial Industrial Accident Exposure and Social Vulnerability Assessment of Hazardous Material Sites, Chemical Parks, and Nuclear Power Plants in Germany

Industrial accidents have shown that many people can be affected, such as in Seveso, Italy, in 1976. Industrial accidents in nuclear power plants have also led to fatalities and evacuations. To better guide preparedness against and mitigation of industrial accidents, an assessment is necessary to evaluate hazard exposure and the type of potentially vulnerable social groups that need to be taken into account. This study conducted a spatial assessment of three types of industrial facilities in Germany: facilities registered under the Seveso Directive, chemical parks, and nuclear power plants. The method consisted of a spatial assessment using a Geographic Information System of exposure around hazardous sites registered under the Seveso Directive in Germany and of census data to analyze social vulnerability. Hazards analyzed included industrial accidents and earthquakes. The results revealed that most industrial sites are in urban areas and that population density, the numbers of foreigners, and smaller housing unit sizes are higher in close proximity to these sites. The buffer zones analyzed in circles between 1 and 40 km show a decreasing vulnerability with more distance. This can guide emergency management planners and other stakeholders to better prepare for major accidents and better devise disaster risk reduction strategies specifically for different social groups.


Introduction
In July 1976, an industrial accident near the town of Seveso in northern Italy led to environmental pollution with dioxin, which killed over 2000 animals and led to about 35,000 people affected, 5,000 people evacuated, and 200 chloracne cases among local residents in seven communities (Bertazzi et al. 1998;Pesatori et al. 2003;Nerin et al. 2014). A vapor cloud from an exothermic reaction led to high exposure to tetrachlorodibenzo-p-dioxin. Inhalation, dermal contact, and consumption of contaminated crops mainly led to illnesses and increased cancer risk for the population, but few mortalities of people were clearly related to this event (Eskenazi et al. 2018). The Seveso event received wide attention, especially in Europe, and triggered industrial and environmental safety regulations and assessments, such as the Seveso Directive (Raufer 1996). The aim of the Seveso Directive is the prevention of major accidents involving hazardous materials, with potentially great consequences for environment and population (Delvosalle et al. 2017). Following the Seveso Directive, establishments handling dangerous substances are divided into lower-tier and upper-tier establishments, according to the amount of dangerous substances handled and stored (Delvosalle et al. 2017). Countries such as the UK developed new regulatory measures for the control of major accident hazards (COMAH) (De Cort 1994).
Regarding vulnerability, the Seveso accident was one of the first occasions to observe cancer risk in humans related to dioxin exposure. Especially children under 15 years and women were affected (Pesatori et al. 2003;Eskenazi et al. 2018). The Seveso example showed that industrial hazards affect populations in various ways and different age groups. In other hazard contexts, such as natural hazards research, disaster risk reduction (DRR) promotes risk assessments to 1 3 investigate people's vulnerability more vigorously, not only hazard and damage (United Nations 2015). Therefore, this study analyzed variables to capture social vulnerability in the context of potentially hazardous facilities, including age, gender, and population density.
The risk of exposure of populations to industrial accidents is especially high when residential areas are in close proximity to chemical facilities-residential areas often lie even within a 1 km distance to these facilities (Tahmid et al. 2020). Hazardous waste disposal site studies and related industrial risk studies also use a 1 km buffer (Gómez-Delgado and Tarantola 2006). Another buffer distance used in the Seveso Directive registered sites databases is 5 km (Bogliolo 2012). Exposure assessments that include social vulnerability information often use Geographic Information Systems (GIS) to analyze buffers up to 10 km around industrial sites (Gerbec and Kontic 2002;Ştefănescu et al. 2018). At the original Seveso accident site at the ICMESA industrial plant in Italy in 1976, three exposure zones and distances were identified and mapped-Zone A: 3 km, Zone B: 7 km, and Zone R: 9 km (Bertazzi et al. 1998). In Zone R, about 32,000 residents were warned about local food consumption (Eskenazi et al. 2018). Spatial dispersion footprints are often modeled as circles, ellipses, or cones, representing thermal radiation, aerial transport, or other (Tahmid et al. 2020). Spatial ranges depend on several factors, mainly including the amount of hazardous material, explosiveness, wind speeds, and wind directions (Lees 2012).
Spatial planning tools using GIS are applied in analyzing industrial risks and for Seveso site information systems (Kirchsteiger et al. 1999). The ongoing growth of industrial sites in the European Union underlines the need for such assessments (Török et al. 2020). Many funded research projects typically involve assessments of the likelihood of accidents, the vulnerability of the surroundings, and safety barriers and processes (Salvi and Debray 2006). Traditional spatial risk assessments have long focused on the hazard aspects only, such as dangerous substances (Bubbico et al. 2004). Increasingly, vulnerability is analyzed, but mainly in terms of land use and environmental impacts or regarding exposure of built-up areas. Many tools and decision support systems have been developed that typically include an accident database, a database of dangerous substances, and spatial overviews of land use and potential environmental vulnerabilities (Orso Giacone et al. 2007).
Decision-support systems are quite common in the area of industrial accidents or Natech (Natural hazards triggering technological accidents) (Suarez-Paba et al. 2019;Suarez-Paba et al. 2020), and have also been using GIS for some time (Chang et al. 1997;Zografos et al. 2000;Luo et al. 2022). These decision-support systems can also include elements such as hospitals, roads, and rivers that are potentially exposed or population density Bonvicini et al. 2012;Zhao and Liu 2016;Ştefănescu et al. 2018). However, in many cases, they do not contain information about the social vulnerability and more details about the population (Tahmid et al. 2020), as is the case in natural hazard contexts (Cutter et al. 2003). Social vulnerability assessments analyze distinct social vulnerability classes-age, gender, income, and other factors-and can be combined with environmental vulnerability and hazard control information (Radosavljević et al. 2018). Some existing social vulnerability indicators and indices developed in natural hazard contexts are applied to industrial risks, especially nuclear power plants (NPP) (Kyne and Harris 2015;Pence et al. 2019). Risk, as a combination of hazard exposure and vulnerability, is understood as information for decision makers to prioritize, for example, in emergency management (Birkmann 2013). More recent and advanced assessments combine GIS assessments of spatial dispersion footprints using ALOHA software. Such spatial dispersion can then, for example, be combined with a composite index of social vulnerability indicators such as age, gender, and other aspects (Tahmid et al. 2020).
While this study focused on Germany and literature mainly on Europe as a broader context, many industrial hazards and accidents also occur in low-middle-income countries (Jasanoff 1994). Related research has analyzed industrial contaminated sites (Papu-Zamxaka et al. 2010;De Sario et al. 2018), hazardous waste accidents in Africa similar to Seveso (Harjula 2006), or major chemical accidents in Latin America (de Souza Porto and de Freitas 2020).
The risk concerning chemical and industrial sites persists in Europe or is even increasing (Wood and Fabbri 2019;Hollá et al. 2021). On average, around 30 accidents are reported for the EU per year, and according to statistics, it is expected that every hazardous facility will experience an accident during its operational life (Kirchsteiger 2001;Wood and Fabbri 2019). Within the European Union, Germany has a high concentration of potentially hazardous facilities (Suffo et al. 2015). However, no studies were found in a literature search that cover all of the country's Seveso sites, chemical or hazardous materials, and capture both spatial exposure and vulnerability in several variables, or compare them to nuclear power plant site risks. Therefore, this study was aimed at analyzing risk in Germany with a spatial assessment approach. The reason is that Germany is a major industrialized country with many industrial sites. But what is missing so far is a spatial overview of the risk associated with these sites and how it could interact with the residential population. Past accidents have shown that populations are unequally affected, with women and children especially prone to detrimental effects. This article is organized into sections that first describe the methodology, then the results of the GIS analysis, and finally discuss the results regarding the observations on vulnerability, exposure ranges, and differences in distances to the sites. To get a better feeling for the results, not only Seveso sites were analyzed, but also potentially hazardous sites in Germany, including chemical parks. The results were also compared to NPP sites. The goal was to inform science and administrative agencies monitoring such risks about spatial patterns and the options available using GIS. To guide fellow scientists and users, the results are also discussed in terms of the observed constraints and limitations of the approach to contribute to a better and more transparent understanding.

Methods
The method used consists of a place-based social vulnerability approach that identifies risks by hazard exposure and vulnerability components. It is known that capacities and resilience are important additional factors to consider (Bertazzi et al. 1998). However, no such information was accessible through public sources. The literature search was conducted systematically by selecting specified search terms and adjusting them according to the search results. The main body of literature was retrieved from Scopus and Google Scholar with the search terms: Seveso AND exposure OR risk OR vulnerab*; Seveso GIS exposure, Seveso GIS exposure Germany.
An open-source geographic information system, QGIS, was used to enable overlay and numerical analysis of an aggregation of spatial exposure areas with underlying population and demographic information. The distances of the radii are in accordance with the literature. While Seveso site studies and related hazardous material studies often do not go beyond a 10 km range, a 20 km and 40 km radius was added in the study. The reasons for this are studies in neighboring areas, such as on nuclear accidents that use wider ranges of up to 80 km distance or more (Susnik 1987;Mohammed Saeed et al. 2020). Another argument is that this study intended to analyze spatial vulnerability patterns resulting from existing hazardous sites. A potentially hazardous facility is perceived as a risk by residents beyond a 10 km distance (Delvosalle et al. 2017). Risk perception studies would be an interesting and important addition, but the main purpose of this study was to identify whether spatial vulnerability patterns change with growing distance. Circular buffers were selected since wind directions can change. While Germany is located in the northern west wind zone, and highs and lows basically move from west to east, the meandering wind regimes result in wind from all directions. Based on the real case of the Seveso accident in 1976, distances between 3 and 7 km seem realistic (Bertazzi et al. 1998).
The vulnerability variables were selected based on the literature. However, another underlying reason is the limited availability of open-access data on demographic and census information in Germany. Because of privacy and data protection rules, as well as the federal system, nationwide data are available mostly for administrative areas only. However, the most recent 2011 census data are available at a 1 km grid resolution for a few selected demographic and housing variables. Data on some common social vulnerability indicators, such as the educational level and household income, are unavailable. All available variables were included in this study. However, some compromises had to be made. For example, children under 15 years are identified in the Seveso accident as especially vulnerable (Pesatori et al. 2003;Eskenazi et al. 2018). The only similar data provided for Germany are children under 18 years of age. Gender ratio is available, as well as population density. This selection of additional variables is based on their usage in social vulnerability assessments in natural hazards research (Cutter et al. 2003;Birkmann 2013). For example, foreigners were selected because they are often characterized by language problems that can hinder the understanding of warnings about hazards or ongoing events (Cutter et al. 2003). Variables on housing were also added, including household sizes, living area sizes per person and housing unit, as well as vacancy of housing units. This kind of housing information provides indirect information about the economic vulnerability of residents. Greater living area is correlated with higher income. Vacancy of housing units is a proxy for the economic profile of a community; unoccupied housing units relate to less demand on the market (Kabisch 2005;Wiechmann and Pallagst 2012). It is not a perfect proxy for regional economic power, but in the absence of other data, a high vacancy of housing units can be interpreted as lower economic power compared to other regions. With regard to potential biases in our initial dataset we checked for the following: not all women or all elderly citizens are automatically more vulnerable, but for a whole society such as a country, on average, these effects have been measured in several hazard and country contexts (de Sherbinin 2014; Rufat et al. 2015). Other known limitations are that indices often blur the more specific individual indicators (King 2001). Therefore, we also show the indicator ranges in a table and discuss individual indicators with the respective maps and example areas.
The census data from 2011 were collected and the methodology documented and checked by the German Federal Statistical Office (Statistische Ämter des Bundes und der Länder 2020). The census data information was checked for inaccuracies using the open statistics software PSPP. The detailed steps of the assessment of spatial industrial accident exposure and social vulnerability in Germany are summarized in Fig. 1.
In the methodological steps, first potentially hazardous sites were mapped from various sources; the Seveso site coordinates were obtained from the German Environment Agency (UBA/Umweltbundesamt, personal 1 3 communication 1 ), the chemical parks were mapped from an online map of the chemical industry association of Germany, VCI (Verband der Chemischen Industrie), 2 and the nuclear power plants (NPP) were mapped from the Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection website. 3 The locations of the chemical parks and NPP were then adjusted for more precision using OpenStreetMap. The Seveso sites were checked for accuracy, and a few addresses were corrected from a list of facilities in Germany provided by the German Environment Agency that oversees all facilities that involve hazardous materials and waste. Only upper-tier establishments of sites under the Seveso Directive were considered in this study, since the quantity of dangerous substances is higher than in lower-tier establishments.
In the second step, demographic data retrieved from the 2011 census were analyzed and cleaned. Areas with values of -1 and -9 were deleted, as they designate non-populated or classified areas. Each category was analyzed for mean values. Zero values were kept as zero values, as they indicate that zero observations were made. Comparing two available datasets, we decided to select all census data as point data for the study, not grid data. The reason is that the census adopted the more conservative approach, with more spatial precision.
The data were loaded into QGIS, assigning x/y coordinates for the point data, using a UTM projection model, zone 30N with ETRS89 and EPSG25832. In the third step, hazard zones were created with buffers of 1, 5, 10, 20, and 40 km around Seveso sites, NPP sites, and chemical parks. The circular buffers were clipped with the census data to analyze the social vulnerability spatial distribution.

Results
The results of the spatial assessment of the risk of chemical plants and hazardous industrial sites that affect the residential population are shown in maps and tables. The spatial assessment of risk is also analyzed for NPP to enable comparison and complement the picture of industrial risks. In accordance with the risk definition used, exposure is shown first and then vulnerability. Accidents triggered by a natural hazard or disaster that result in the release of hazardous materials are commonly referred to as Natech (UNDRR-APSTAAG 2020). Natech risk management includes "understanding the nature of cascading risk these natural hazards can pose to different infrastructure and services and thereby having a potential to become technological disaster" (UNDRR-APSTAAG 2020, p. 15). As an example of a natural hazard, earthquake exposure in Germany is shown in the last subsection. Due to the convenience of common technical language, chemical sites, industrial plants, and waste locations are summarized under the term "Seveso site" (Delvosalle et al. 2017). "Seveso plant" might be more common, but our data includes industrial plants, landfills, and other facilities and locations, registered under the Seveso Directive in Germany. What is termed human or population vulnerability in some literature identified was summarized under the term social vulnerability.

Population Exposure to Seveso Directive Registered Sites
The resulting map of hazardous sites in Germany (Fig. 2) shows a large number of such facilities. A large segment of Germany's population is exposed to risks from hazardous substances. As calculated from the population density data of the census, approximately 6 million people are living within a 1 km distance, 34 million are living within a 5 km distance, and 54 million people are Fig. 1 Methodological steps of the assessment of spatial industrial accident exposure and social vulnerability in Germany. OSM Open-StreetMap, UBA Umweltbundesamt / German Environment Agency, VCI Verband der Chemischen Industrie / Chemical Industry Association, Germany 1 Since the export of the data about Seveso sites publicly available online in the eSPIRS Database (https:// espirs. jrc. ec. europa. eu/ en/ espirs/ conte nt) was not possible, an information request to the German Environment Agency was sent via email. A dataset, updated on 30 June 2021, was provided by an employee of the German Environment Agency, Department of Sustainable Production, Resource Conservation and Material Cycles and sub-division of plant safety via email on 31 January 2022. 2 https:// www. vci. de/ vci/ downl oads-vci/ bilder/ stand ortka rte-zahlen-300dpi. jpg 3 The locations were taken from this website and additional data were collected based on web-based search: https:// www. jodbl ockade. de/ anspr echpa rtner/ kernk raftw erke/. living within a 10 km distance to those chemical sites that are recorded by the federal environment agency under the Seveso Directive.

Social Vulnerability Index Overlay
Aggregated social vulnerability for German counties is shown in darker colors for areas with higher potential social vulnerability (Fig. 3). The aggregated social vulnerability is based on previous studies that developed a composite social vulnerability index (SVI) using census data at county level (Fekete 2019). The purpose here is to show how in principle such an existing index can be used for comparison with other hazards such as the Seveso sites. The method included factor analysis, principal component analysis with regression, and validation by an independent second dataset (Fekete 2009), updated with census data from 2015. It was originally developed for natural hazard contexts and lacks specifics for chemical and industrial hazard contexts. However, certain variables of social vulnerability are found to be useful in several hazard contexts, and are considered hazardindependent (Schneiderbauer and Ehrlich 2006). Therefore variables such as age, gender, and foreigners were selected as individual social vulnerability indicators, as explained in the methods section.
The overlay of the Seveso sites, chemical parks, and NPP shows an uneven distribution. In the center, and especially to the east of the center in Germany, there is a high ratio of counties that see a combination of vulnerability factors, such as old age, low income, and community debts. Seveso sites are common in those areas. This is due to the history of industrial production in the former German Democratic Republic. The distribution of the other facilities in other areas of Germany provides no distinct spatial pattern at first glance.

Individual Social Vulnerability Indicators
Another dataset than that used in Fig. 3 was used to analyze social vulnerability in more detail. The descriptive statistics reveal missing values that are unevenly distributed among the variables (Table 1). As is commonly the case with such datasets, in this specific dataset the missing values may be connected to privacy, economic, or other ethical concerns. The descriptive statistics (Table 1) display the remaining sample sizes (N), average, minimum, and maximum values, and standard deviation per variable in the census data. The exposure radii in distances from 1 km to 40 km were merged with the census data from 2011. The mean values per exposure radius class were calculated, and ratios and percentages of women, age, foreigners, and housing factors are shown in Table 2.
The variables were checked statistically on significance using a non-parametric Friedman test, which rejects the hypothesis that there is no correlation ( Table 3). The asymptotic significances were high in all three groups (p < 0.001)-Seveso sites, chemical parks, and NPP. It can be inferred from these results that the ranks of increasing distances from the potentially hazardous sites are not random. However, we suggest not aggregating these variables without further statistical testing and removing highly correlated variables. The main purpose of our study is to show how typical variables for social vulnerability indicators, such as foreigners or age, can be used in principle.  Overall, the analysis reveals that some demographic variables show a distinct trend that is visible across the increasing distances from Seveso sites and chemical parks especially. For NPP, fewer correlations are visible. The clearest picture is indicated by decreasing average population densities for the Seveso sites and the chemical parks. From a population density of over 1,000 inhabitants per km 2 , it decreases to around 400 people per km 2 . That means more people live closer to such facilities or in their direct surroundings than at a greater distance. The next clear indication is foreigners, also with decreasing intensity around Seveso sites. The same trend is visible for chemical parks and NPP but not as distinct as with the Seveso sites. The third clearest parameter is area per housing unit, which continuously increases with a clear signal for the Seveso sites, from around 89 to 103 m 2 per housing unit. The same pattern can be observed for settlements around chemical parks; however, the signal from the data is less clear.
Other patterns are also observable from the data, but since the quantitative distance between results is much lower, they should be treated more cautiously. See more about this in the discussion section. The number of women decreases around Seveso sites from 5 to 20 km distance. The ratio of women also decreases from 5 to 40 km around chemical parks. The average age increases around Seveso sites from 1 to 10 km distance. Children and youngsters under 18 years of age increase around Seveso sites from 1 to 10 km. However, the numbers decrease around chemical parks from 1 to 20 km distance. The number of citizens over 65 years of age decreases around Seveso sites from 5 to 20 km. They also decrease consistently around chemical parks from 1 to 40 km distance. Household sizes increase continuously around Seveso sites from 1 to 20 km distance. There is no clear signal for chemical parks, but the number of persons per household increases from 1 to 20 km distance around NPP. Vacancy of housing units decreases around Seveso sites from 1 to 20 km. Around NPP, vacancies also decrease from 5 to 40 km distance. The living area per person in an housing unit increases continuously for Seveso sites from 1 to 20 km distance. It also increases around chemical parks from 5 to 40 km distance.
Overall, the clearest results are for three categories-population density, foreigners, and housing unit size. The spatial patterns are also consistent for chemical parks except for three categories-average age, household size, and vacancy of housing units. The clearest signal here on average is again population density. Only three variables show a direction for NPP, but with much less clarity than around the other sites. The clearest signal is for foreigners, followed by household size, and then vacancy of housing units. Comparing the three potentially hazardous facilities types, Seveso sites have the clearest signal in the data and differences between Housing unit size (m 2 ) 9.00 9.00 9.00 populations in growing distance to the sites, followed by chemical parks.
The results can also be compared to Germany's average of all demographics and housing variables (see Table 2, bottom row). This can help characterize whether industrial sites in Germany are generally located within remote and unpopulated areas or, rather, in densely populated areas. This can also help studies later compare Germany with other countries and hazardous facilities. For almost all data results for all three types of sites analyzed, the average population density is higher than the German national average. One exception is the 1 km distance around NPP. Regarding women, the signal is not so clear, but in all cases except one, females are more common in the proximity of the three site types than in the national average of Germany. The one exception is the 10 km distance around NPP. The average age signal is not so clear and quite similar to the national average. The same is true for children under 18 years of age. For citizens over 65, the ratio is higher than the German average. For foreigners, the ratio is always higher around the sites than the German average. There is no clear picture for household size and vacant housing units. For living area per person, the ratios are only higher for NPP sites at distances from 5 to 40 km as compared to the German average. The housing unit size is the sole parameter that deviates from the German average in both directions, lower and higher. The average housing unit size is lower than the German average for all Seveso sites and all chemical parks. It is higher around NPP for all distances.

Foreigners as Social Vulnerability Proxy
Looking in more detail into the spatial distribution of social vulnerability, the density of the foreigner population was selected. The density of foreigners around hazardous sites shows concentrations in western Germany and Berlin and some other larger cities in eastern Germany (Fig. 4).
The distribution of foreigners is more concentrated along the western border, as identified in another recent study (Fekete 2022). The vulnerability information presented here can be used by individual communities to identify where their concentrations of especially vulnerable people or people with specific needs, such as language and translation for foreigners, are located. Such overviews, as shown here, can also be used by administrative agencies to identify a general concentration of locations and for planning future sites that could avoid such exposed areas. The other social vulnerability variables analyzed are not shown in more detail. That is because population density is nearly coherent with the distribution of foreigners. While population density also points towards socioeconomic conditions, typically associated with urban centers, it is also correlated with many other factors. It, therefore, can guide emergency management mainly toward identifying dense population pockets and numbers of people to taken into account in general. In some studies, other more detailed social vulnerability variables have been more effective in explaining risk than physical vulnerability variables (Fatemi et al. 2017;Ardalan et al. 2019). The other variables in this study, such as gender, and young or old people, are relevant for social vulnerability, too. However, since the signal in the data is not strong enough, they are not mapped here. For this reason, no aggregated vulnerability map or risk index was computed. This is due to the low number of only three variables that show a strong signal. However, it is also due to the critique of aggregated indices (Rygel et al. 2006;Schmidtlein et al. 2008). Individual indicator maps can be helpful to guide stakeholders in addition to aggregated indices but foreigners are not per se more vulnerable than other people. Some studies on floods in areas such as Cologne have found that networks among foreigners are stronger than among wealthy people in the same area (Welle et al. 2014). Therefore, like other vulnerable groups, foreigners have both positive and negative indications of social vulnerability. This is another argument for showing foreigners by individual indicators and maps and not by indices in this case.

Exposure to Earthquake
There are only few maps available for natural hazards and human-made risks at the national scale in Germany. The earthquake example based on DIN EN 1998-1 is illustrative since earthquake risk in Germany is high in distinct regional areas in the west, south, and a pocket in the east, though it is lower than in many other areas worldwide, and casualties are not common. Still, earthquake shocks can trigger environmental pollution from hazardous facilities. Such facilities exist even in the highest earthquake risk zone west of Cologne (Köln) and south of Stuttgart. Figure 5 shows the distribution of Seveso sites with a 10 km buffer and reveals that many of these sites are exposed to earthquake risk.

Discussion
The main purpose of this study is of indicative nature. The intention is to showcase how spatial data can be used in combination with hazard and vulnerability analysis. Especially the conceptual components of spatial exposure around hazardous sites in combination with social vulnerability need more investigation and will hence be discussed a bit more in this section.

Exposure
The exposure information shows that there is a difference between industrial sites. A large number of Seveso sites is correlated with high urban density in Germany. This is also the case for chemical parks and NPP, but they also include locations with lower population densities. Nuclear power plants were designed and planned for areas with lower populations to reduce risk. However, the potential of greater casualties is still not to be underestimated for NPP because of the hazardous substances and the potential for the broad spatial distribution of high risk areas in case of a nuclear accident, as observed for Chernobyl and Fukushima (Steinhauser et al. 2014).
Interpreting the vulnerability data, there appears to be a higher general exposure due to higher ratios of people around Seveso sites and chemical parks. This is also the case for NPP, as compared to the average in Germany (except for the 1 km buffer). The next clear predictor seems to be the ratio of foreigners, which is higher and decreases with distance to all three sites. This means that in case of a large accident, more people need to be rescued, treated, or evacuated closer to such sites (Zhao et al. 2018). Moreover, more emergency management measures must be prepared and adjusted in different languages and according to the different needs of foreigners or people speaking other languages. The third clearest parameter is housing unit size. Since there are smaller housing units in closer proximity to chemical sites and nuclear sites, with decreasing distance, it can be estimated that lower incomes and less space for shelter or long-term in-house sheltering can be assumed. For emergency management, this indicates that more efforts and resources need to be devised for areas at short distances of 1 to 10 km from industrial sites. This also includes economic recovery, financial means for insurance or health treatment costs, and other factors associated with housing, living conditions, and economic profiles.
The constraints of our study are that differences in the day-and nighttime populations are not captured, as well as outdoor places or mobile persons such as travelers (Bonvicini et al. 2012). Furthermore, since hazard parameters for the sites are not differentiated, such as the size of the industrial plant or the stored volume of substances, exposure radii may be overestimated. However, statistics show that exposure up to 30 km has occurred for nonnuclear sites as well (Lees 2012). While the other parameters also display tendencies, they must be treated more cautiously.

Spatial Risk Patterns of Social Vulnerability
Another important aspect that needs further discussion and analysis is the distribution of the population and the relation to risk or potentially hazardous facilities. The exposure and vulnerability assessment carried out here can only indicate correlation but not causality between the location of residential areas and hazardous sites. In the history of major industrial accidents, Seveso 1976 and Bhopal 1984 or the Buncefield fire 2005, stand for disasters that have affected the immediate residential population around it. There were explosions in production facilities before World War II and also some accidents in chemical parks or at major industrial sites that led to environmental pollution, for example, Schweizer Halle 1986 (Giger 2007). While such events have attracted greater media attention and have led to enforced environmental pollution regulations, fortunately, no similar event to the Seveso or Bhopal event has occurred in Germany after World War II. This means that it is difficult to validate how people have been affected and whether vulnerability profiles of age or gender have made a difference. However, validation studies are increasingly being demanded (Rufat et al. 2019), and some exist for chemical accidents (Ardalan et al. 2019).
However, another aspect often less discussed in vulnerability studies is the problem of correlating residential areas with risk. It could be assumed that people would react to the risk of potentially hazardous facilities in their neighborhood, resulting in poorer populations around such sites. Such patterns have been observed in the United States with respect to NPP (Kyne and Harris 2015;Pence et al. 2019). In Bhopal, India, the poor segment of society, the local workforce, was mainly affected (Dhara and Dhara 2002). In Germany, it appears that most of the Seveso sites are located in urban areas. Population density is much higher in the direct vicinity of those sites, as the data in the study show. However, the data do not reveal whether this concentration is due to the existence of those sites or not. This is also not the purpose of this study. Our assessment aimed to analyze how many people are exposed and whether their vulnerability pattern provides signals to better guide emergency management measures in case of a major accident. It can also help to direct information and communication campaigns to deal with minor events such as vapor clouds 1 3 that occur more frequently. Foreigners and elderly citizens have to receive information and preparedness advice differently due to problems of understanding or mobility (Cutter et al. 2003;Wisner et al. 2004;Birkmann 2013). Women and children need to receive different medical and health precautions and treatments.
The spatial distribution of residents in Germany is due to several factors. Urban areas are attractive because they provide jobs, and the chemical industry and other industrial facilities such as NPP are important employers for many people. This may explain part of the population density around such sites. Based on previous studies on risk perception in Germany, chemical and nuclear accidents have greatly shaped public perception in Germany (Renn 1990;Beck 1992). Local studies integrate the perception of the severity of such risk into vulnerability indices (Glatron and Beck 2008). Risk perception can motivate people differently to avoid or take risks for job opportunities and proximity to urban centers (Slovic 1987;Huang et al. 2010). In other countries worldwide, the social vulnerability in developing countries, for example, is often strongly correlated with poor people who have no choice but to choose hazardous environments to be close to cities and job opportunities (Wisner et al. 2004). It is, therefore, interesting and important to analyze whether the same pattern exists in industrialized countries such as Germany. The study subject warrants further research, also using statistical methods for testing the original and processed data and validation studies. For instance, further study is needed to establish an aggregated index and other (for example, spatial) correlation methods. While the study allows for no causality correlation, it does show that higher population density, foreigners, and smaller housing unit sizes create a typical risk pattern. This can guide stakeholders in different fields, such as urban planning, environmental planning, or emergency planning, in future strategies for dealing with spatial risks. While most studies still use information about the hazard and exposure by location of land-use and buildings only, the study also points towards opportunities to utilize more information from census data to get a better understanding of different vulnerabilities of people.

Conclusion
Many guidelines developed after major chemical or industrial accidents call for risk assessment. Increasingly, such assessments consist not only of hazard information and physical exposure. Social vulnerability parameters increasingly enrich such risk assessments and are demanded in a context such as natural hazards. International agendas by the United Nations, such as the Sendai Framework for Disaster Risk Reduction 2015−2030, call for more coherent risk assessments that consider vulnerability and integrate perspectives on different types of human-made and natural hazards (United Nations 2015).
This study has therefore conducted a spatial exposure and social vulnerability assessment. The results reveal that the vulnerability of residents in the vicinity of industrial sites decreases with growing distance. The data show that population density decreases as well as the number of foreigners around those sites, and housing unit sizes are smaller in close proximity to those sites. The results also show that the risk varies between Seveso (directive) sites, chemical parks, and NPP sites in Germany. Seveso sites are most correlated with urban areas. The results can inform planners in different fields of environmental, urban, and disaster risk management on how to prepare for major accidents. Evacuation, information and communication activities, sheltering, decontamination, health treatment, insurance, and economic recovery can be more customized based on such information. Limitations of this study lie in data availability, especially for social vulnerability, where the census data in high spatial resolution is currently still very much restricted to only a few available variables. The study does not evaluate risk in all its components, only exposure and social vulnerability, but not different degrees of hazard magnitudes or frequencies other than buffer zone exposure scenarios. However, the spatial assessment provides an important opportunity to assess such risks over larger regional areas and update and monitor them continuously. Such monitoring has been triggered and streamlined by the United Nations (UNISDR 2017). These initiatives call for data collection not only of people affected but also of other details such as vulnerability and exposure information. This study, therefore, informs about data and methodology opportunities to report and monitor such risks.