In the context of knowledge sharing, the term ontology is used to mean a formal specification of shared knowledge so that vast amounts of information, data, and concepts can be structured and organized for storage, querying, and retrieval (Gruber 1993). The structure provided by the ontology can be understood as a formal, hierarchical representation of concepts and their interrelations in a specific knowledge domain (Raskin and Pan 2005; Mainz et al. 2008). Common components of ontologies are individuals, instances or objects (the basis or “ground level” objects), classes (sets, collections, concepts, classes in programming, types of objects, or kinds of things), properties (aspects, attributes, features, characteristics or parameters of the objects and classes), and relations (ways in which classes and individuals can be related to each other). Once developed, this abstract structure enables the user to depict the structure of its knowledge domain by collecting synonyms, capturing hierarchies like in taxonomies, and establishing relations between classes and individuals (Mainz et al. 2008).
The ontology illustrated in Fig. 2 shows that each vulnerability assessment is the basic object or Instance of the ontology, which belongs to a Category X. In the later implementation in the semantic wiki, each wiki article also is an instance of a Class C and has a Value V for Property P. Accordingly, the Instance: Vulnerability Study of Electrical Systems for Category: Vulnerable System can be classified to have Class: Technical System that has Property: Infrastructure with Value: Electrical System.
Four key questions form the first level “branches” or categories of the ontology and correspond to the basic, abstract structure of the knowledge domain of vulnerability assessments and, hence, to the entry point of the semantic wiki later on. The four questions are simple, yet consequential questions, and have been deduced from various theories and concepts from a multitude of disciplines: (1) Vulnerability of what? (2) Vulnerability to what? (3) What reference framework was used in the vulnerability assessment? and (4) What methodological approach was used in the vulnerability assessment? In the following subsections, we introduce and explain the ontology along these four basic questions. While reading them, it is important to keep in mind that when describing the ontology we use language in a formal way independent of technical jargon in the field and regardless of the fact that in some cases the same word (for example, “driver”) might also be used in a certain vulnerability concept with a specific meaning. To avoid misunderstandings, it is therefore important to distinguish between the formal level of the ontology and the content of the knowledge domain described in the ontology.
Vulnerable Systems—Vulnerability of What?
To answer the question “Vulnerability of what,” we use a systems approach (system as a collection of parts or subsystems) and begin with the classic concept of “risk” as we find it in natural science or engineering domains: risk is a function of hazard and vulnerability. While the hazard is commonly referred to as the occurrence potential of a triggering event, the notion of vulnerability designates the predisposition of people, processes, infrastructure, services, organizations, or systems to be affected, damaged, or destroyed by the event. In this concept, hazard is the exogenous and vulnerability is the endogenous variable of risk. Something is at risk, exposed to or affected by an occurrence (perturbation, stress) and something possesses the potential to change its state, a degree of sensitivity, and the capacity of response. This quality exists a priori. In general, the object of observation is thought of in abstract terms as a system. In developing the ontology, it is therefore assumed that every research into vulnerability must imply the distinction of system and environment and must, furthermore, distinguish types of systems and subsystems investigated in the study, as this is the most basic premise in general systems theory (von Bertalanffy 1950). Consequently, the ontology on vulnerable systems shown in Fig. 3 explicitly refers to four classes of vulnerable systems: (1) “natural systems” for vulnerability studies referring to a set of subclasses that include physical systems (Calvalieri et al. 2012), biological systems (De Lange et al. 2010), and/or biophysical systems (O’Brien et al. 2004); (2) “social systems” for vulnerability studies referring to the subclasses of population in general (Adger 1999; Carreño et al. 2007), social groups, for example, communities (Cutter et al. 2003; Bollin and Hidajat 2006), functional systems, such as the economy (Patt et al. 2010), the public financial sector (Mechler et al. 2006) or the health sector (Hahn et al. 2009; Few and Tran 2010); and (3) “technical systems,” such as vulnerability studies referring to critical infrastructure (Hellström 2007; Kröger and Zio 2011). In addition, the ontology also accounts for a separate class of hybrid concepts referring to interactions between and within systems, such as in societal and ecological (biophysical) subsystems (Turner et al. 2003; Gallopín 2006) or societal and technical subsystems (Khazai et al. 2013).
Overall, the ontology on vulnerable systems shown in Fig. 3 mirrors some classic approaches of hazard and vulnerability research, but also includes sociological theory in the form of a strict distinction between modes of operation of natural, social, and technical systems as well as the thesis of functional differentiation of modern society (Luhmann 1997). The strictness of this argument (functional differentiation) could not be maintained in some cases, since it would have left out or pushed aside established nomenclature in mainstream vulnerability research. For example, the branches of “Industry,” “Agriculture & Forestry,” or “Financial System” are certainly part of the overall economic system, but in most studies they are referred to as complementary systems on their own. In recent years, attempts were made to introduce sociological terminologies in vulnerability research (Zehetmair 2012). It remains to be seen to what extent these attempts will meet with acceptance.
Vulnerability Drivers—Vulnerability to What?
One of the basic traits of the concept of vulnerability is the need to analyze the relationship between system and environment regarding contingent occurrences (shock) or rather slowly developing changes leading to unsafe conditions (continuous stressors). However, there are many nuances in the nature of the correlation between hazard and vulnerability. By asserting that “hazard and vulnerability are mutually conditioning situations and neither can exist on its own,” Cardona (2003) raises awareness towards conceptual issues with the a priori existence of hazard and vulnerability separate from each other. Therefore, it is important to highlight the theoretical model behind the vulnerability analysis.
The dominant concept in vulnerability research is that of factor-theoretical models of an explanation of cause and effect relationships, which refer to the idea of “causality.” In our ontology, the term “driver” was chosen as an abstract term to answer the second basic question: Vulnerability to what? In the ontology, “driver” refers to instantaneous events and/or long-term processes as well as to external and/or internal causes. Among many other features, general systems theory claims that systems maintain contact to their respective environment in a very selective fashion, despite sustaining a boundary between the system and its environment. In terms of causality, the arguments contend that in sustaining a boundary, systems cut off many causalities, while simultaneously they must control some, but not all, causalities vital for their reproduction (Luhmann 1995). Those productive causes must be employed to some extent within the system (as endogenous factors), while others remain environmental causes (as exogenous factors). In this sense, potentially hazardous effects on the system must be defined as unproductive causes that can occur either outside of (external) or inside the system (internal). “Driver” in our ontology therefore indicates how a triggering event or process can influence, affect, or deviate the stability/equilibrium of a system, that is, establish the conditions for maintenance of physical structures or the reproduction of living systems. Instead of discussing “negative” and “positive” effects, which is, as a judgment, always observer-related, we can distinguish in a more abstract way a driver as a productive or unproductive cause related to the system in focus.
For further structuring the “driver” in the ontology, we chose the classes “natural” and “social” drivers. Typical drivers in natural hazard research, which act outside of a system, are called “natural drivers.” In the ontology shown in Fig. 4, they are further subdivided into three subclasses: geophysical drivers (earthquakes, volcanic eruption, landslides, tsunamis), hydrometeorological drivers (tropical cyclones, tornados, floods, coastal storm surges, droughts, and so on), or biological/ecological drivers (for example, infestation or loss of biodiversity). Whether a system is vulnerable to processes of endogenous risk production (self-endangering) of a system itself is of importance. While, for example, from the perspective of the field of engineering the dominating canon of vulnerability assessments is concerned with exogenous “natural” hazards (for example, the vulnerability of a building to an earthquake), studies that analyze vulnerability from a societal perspective focus on endogenous processes of “social drivers” of vulnerability. Those assessments typically cover social inequalities, political systems, and policies as drivers (for example, Pelling 2003; Brooks et al. 2005; Wisner 2006; Hahn et al. 2009) or they concentrate on how decision-making processes contribute to creating vulnerability, like in economics (for example, Smithson 1993). Our ontology tries to integrate classic features of vulnerability research, while remaining open to recent theoretical developments that may be implemented in vulnerability assessments in the near future. Next to the “natural driver,” we attribute considerable importance to the “social driver” and identify social inequality, governance, war and conflict, and anthropogenic impact as different subclasses of drivers within the “social driver” class (Fig. 4).
Furthermore, the conceptual decisions on the “properties” of the vulnerability drivers should be made clear in every study by using a temporal scope of the drivers (observed as a continuous stressor or discrete shock), spatial scope of the driver (local, regional or global impacts), and, in case of hybrid events where there is more than one driver, the interaction between different drivers (for example, cascading and linked hazards) (Fig. 4).
The framework of reference of all vulnerability studies correlates with the answers to the core questions of “Vulnerability of what?” and “Vulnerability to what?” In general, we distinguish three dimensions of assessment—factual (and more specific, spatial), temporal, and social—when describing the reference framework of vulnerability studies (Fig. 5). In this way, the assessments differ in regard to the scope of assessment in the social dimension (individuals: Adger 1999; households: Turner et al. 2003; Eriksen and Silva 2009; communities: Bollin and Hidajat 2006; Wisner 2006), the spatial dimension (region: Ranci and Migliavacca 2010; country: Brooks et al. 2005; subcity: Armas 2008), and the temporal dimension (point of time: Kienberger et al. 2009; medium term: Hahn et al. 2009; long term: Li et al. 2010).
An additional class, the “target users” (for example, scientists, policy makers, local authorities, emergency managers, insurance companies) for whom the vulnerability assessment is made is also described in each study. It is an additional class of the reference framework in our ontology. Each class varies regarding the scope of assessment with which researchers operate. To better illustrate some of the distinctions used in the ontology, three examples are presented below for each of the three dimensions (spatial, temporal, and social) used in the reference framework.
Example 1: Fact/Spatial Dimension in Vulnerability of Critical Infrastructures
The fact dimension of the vulnerable system in which the spatial aspect is the most important specification refers to the distinction of elements within the system and to spatial distinctions, for example, the spatial realization of interrelated elements. While the spatial dimension of the vulnerability of geographical or political units or entities might be rather simple and the focus of vulnerability analyses might be cities (Pelling 2003; Prasad et al. 2009), regions (Birkmann et al. 2012) or whole nations (Birkmann et al. 2011; GAO 2011; Welle et al. 2012), the situation becomes more complex when the vulnerability of a functional system is assessed. One example is the vulnerability of critical infrastructures, where observers are confronted with the fact that physical installations or communication networks are spread out in a distinctive manner. We find systems, networks of systems, or internetworks (Edwards et al. 2007). Critical infrastructures encompass locally sited water supply systems (Möderl and Rauch 2011), regionally implemented power grids (Hines 2010), or globally expanding information and communication grids (Hellström 2007). From a methodological point of view, it is very difficult to distinguish sharp boundaries of infrastructure systems, in which technical and social elements are included and interact in a complex manner. Consequently, the analytical framework is somewhat different in each and every study.
Example 2: Temporal Dimension in Vulnerability to Climate Change
The temporal dimension of assessments correlates with the system in focus, but especially with the driver a system is exposed to. Research into vulnerability and climate change exemplifies the need for a distinctive temporal scope of observation. Research in this domain is driven by (at least) two theses: (1) It is widely assumed that climate change and the occurrence of extreme weather events correlate (for example, Kunreuther and Michel-Kerjan 2009). As a consequence, the scope of hazard and vulnerability assessment must include short-term, instantaneous events as well as long-term developments. Scientific estimations of significant changes in the dynamics of the climate system are in the range of decades and centuries (Lenton et al. 2008); (2) Any design and implementation of action plans must also consider distinctive temporal horizons in preparing for immediate threats or for the adaptation to long-term climate change as well as in responding to sudden weather events and using long-term mitigation strategies (Füssel 2007). For example, researchers call for multiple perspectives when analyzing large urban agglomerations: “A resilient community is one that maintains a current information base to understand potential hazards, and is well informed in the preparation and implementation of its future growth and improvement plans” (Prasad et al. 2009, p. 4).
Example 3: Social Dimension in Vulnerability
Since the mid-1970s, research into vulnerability has included the analysis of situations of vulnerable people and vulnerable groups (O’Keefe et al. 1976) and increasingly implemented means of assessing social realities (Hewitt 1983; Blaikie et al. 1994; Bohle et al. 1994; Adger 1999, 2006; Pelling 2003; Wisner et al. 2004). Blaikie et al. (1994) and Wisner et al. (2004), for example, used a set of variables to distinguish root causes, dynamic pressures, and unsafe conditions and generated a generalized description of “being affected” of individuals or social groups (as families, households, neighborhoods, or groups as “the poor” or “migrants”). Linking vulnerability on the micro level (individuals, households, “groups”) to processes and distant root causes on the macro level has been an immense improvement in explanatory power concerning the overall complexity of hazardous situations, yet it is associated with methodological challenges regarding the social dimension of assessing vulnerability. When analyzing the vulnerability of small, concrete social units as the level where vulnerability is revealed, the analysis at the same time refers to the level of the larger, more abstract social units and levels that help shape and propagate dynamic pressures and root causes. These forces, in turn, determine the unsafe conditions on the small social scale, such as the globally operating economy, the development of large urban agglomerations, or the transformation of modern society driven by functional differentiation.
In the end, the scope for empirical research in vulnerability assessments in most cases is related to smaller units, like individuals, households, neighborhoods, and communities. Consequently, we used these levels also as subclasses for the social dimension in the ontology.
The methodological framework domain of the ontology is subdivided into the operational approach and the underlying theoretical concept implemented in vulnerability assessments.
The ontology for operational approaches used in the vulnerability assessment is characterized by the “research design” class: we distinguish between longitudinal, cross-sectional case studies and assessments, which have a strong focus on defining indicators that measure vulnerability. Since indicators are a key element in operationalizing vulnerability assessments and have a strong impact on the validity of the assessment, a special class in the operational approach ontology is dedicated to “indicators” and is used to provide an overview of the actual indicators used in a particular vulnerability study. Sometimes, the choice of indicators is restricted to secondary data provided by official statistics, whereas in other contexts researchers develop ad hoc indicators. This domain of the ontology shown in Fig. 6 provides an overview of all captured aspects of the operational approaches of vulnerability assessments. In addition to the “research design” and “indicator” classes already described, this includes “data collection” and “data analysis” methods. The “data collection” class describes the methods and sources used to gather information about the vulnerability of a certain place or system. The assessments differ in techniques for data collection, such as remote sensing (Eckert et al. 2011), mapping (Boruff et al. 2005; Collins et al. 2009), available socioeconomic data as input for multivariate statistics (Cutter et al. 2000, 2003), focus groups (Brooks et al. 2005) or content analysis (Turner et al. 2003). Methods of inquiry that focus on in-depth understanding of human behavior and its reasons are labeled as “qualitative.” Often, these methods use nonstandardized instruments and rather ask “why” or “how” something happened instead of “where,” “when” or “what.” In some cases, concerned people or stakeholders participate in steps of the research process and the relationship between researchers and interested parties is less or even non-hierarchical. The “data analysis” class describes various methods used to analyze data in the various vulnerability assessments. This includes attributes, such as multivariate statistics (for example, regression analysis, principal component or factor analysis); content analysis; historical or policy analysis; uncertainty treatment; modeling and simulation; spatial analysis; spatial or temporal mapping; and indexing. For the latter, different approaches to aggregating indicators to an index are distinguished: (1) method of weighting indicators (for example, statistical, expert opinion, multi-criteria decision analysis (MCDA)); (2) method of aggregation of indicators (for example, additive, multiplicative, geometric); (3) selection of indicators that are included in index; and (4) accuracy and validity of the approach.
Concepts of vulnerability and the corresponding definitions of vulnerability vary across research domains and determine the choice and design of research instruments. Hence, a discussion of an assessment should always take into account the theoretical framework and the underlying definition of vulnerability. Each conceptual framework can comprise a multitude of factors which determine vulnerability. Unfortunately, these frameworks are incompatible with each other and no overall framework exists. Füssel (2007) argues that terminological confusion mainly results from an unclear distinction between the dimensions “sphere” and “knowledge domain” and proposes a minimal structure to classify the multitude of approaches. Whereas the first dimension “sphere” describes whether a vulnerability factor is considered as internal or external, the second dimension “knowledge domain” distinguishes between socioeconomic and biophysical factors, which can and do overlap. Socioeconomic factors encompass aspects like access to power and resources, social networks as well as policies, international aid, or economic globalization. In comparison, biophysical aspects of vulnerability refer to topography, environmental conditions, land cover or hazards like earthquakes, storm or sea level rise. Based on Füssel’s ideas (Füssel 2007), the vulnerability assessments were classified according to their main conceptual lineage (Fig. 7): (1) Risk hazard approach (Burton et al. 1978; Kates 1985; Hewitt 1997; Füssel 2007); (2) Political economy approach (Adger and Kelly 1999; Pelling 2003); (3) Pressure and release model (Blaikie et al. 1994; Wisner et al. 2004; Rauken and Kelman 2010); (4) Resilience approaches, such as the MCEER Framework for quantifying resilience (Bruneau et al. 2003) and the Bric Model of community resilience (Cutter et al. 2010); and (5) Integrated approaches, such as Cutter’s Hazard of Place model (Cutter 1996), Turner’s Vulnerability Framework (Turner et al. 2003), and the BBC Conceptual Framework (Birkmann 2006, based on Bogardi and Birkmann (2004) and Cardona (2001)). Integrated approaches are not a homogeneous class, but differ from each other in complexity and abstractness of the theoretical concept, hazard conceptualization, and the degree to which they can be made operational. Regarding the definition of vulnerability, the ontology distinguishes whether vulnerability is defined explicitly or implicitly in a vulnerability assessment.
It would be too space-consuming to list in this article all the vulnerability assessments encoded in VuWiki. A brief general overview referring to a limited number of vulnerability studies using the key components of the ontology described in this section is presented in Table 1.