Given the specific nature of its rapid onset, unpredictability, and destruction potential, earthquakes pose a serious threat to cities, placing their assets and inhabitants at risk. Earthquakes do not affect every country, region, city, or even part of a city equally (Wisner et al.
2004). Although developed countries are at risk for their high-valued assets, developing countries are more at-risk regarding fatalities, causalities, and short- and long-term impacts (Pelling
2003; Peduzzi et al.
2009).
Turkey is one of the developing countries highly susceptible to earthquakes. It is the most at-risk country for earthquake disasters worldwide for urban mortality and economic loss (Brecht et al.
2013). One of the most damaging earthquakes in the country, the İzmit earthquake of 1999 (Mw 7.8, also called the Marmara earthquake) caused the death of over 18,000 people and 37,000 people were injured, and about 95,000 houses were severely damaged (Bağcı et al.
2000). Even though it has not been long past 1999, an earthquake hazard is not a remote possibility for the city. A major, Mw ≤ 7.6 event is expected on the Marmara segment of the fault (west) in the next half century with a probability of approximately 50% (Şengör et al.
2005). İzmit City is expected to be affected by an earthquake in this segment.
Under the threat of a probable earthquake, the seismic risk of the city remains a critical issue to be addressed. Although the city has developed since 1999, there is still evidence of needs for improvement regarding the building stock, socioeconomic status of the inhabitants, and the built environment.
This study aimed at vulnerability assessment and mapping to demarcate vulnerable places encompassing diverse components that produce vulnerability at the part of İzmit City with very high seismicity using a data-driven approach, advancing visual communication of the results, and making a crude estimate of the inhabitants that are at the highest risk.
1.1 Disaster Risk and Vulnerability Concept
Disaster risk is a product of co-existence of hazard, exposure, and vulnerability. While hazard and exposure are components of disaster risk that are barely modifiable, vulnerability is. In this regard, there has been a shift from hazards to a vulnerability/resilience paradigm (Haque and Etkin
2006; McEntire
2012) and an increased recognition that disaster risks on human society cannot be reduced without a focus on vulnerability and its quantitative assessment.
Vulnerability is a complex and multifaceted concept (Bohle
2001), and there are many conceptualizations of the term, as gathered in the review by Diaz-Sarachaga and Jato-Espino (
2019). Among many others, a widely recognized definition suggests that vulnerability reflects the susceptibility or the intrinsic predisposition to be affected or the conditions that favor or facilitate damage (Cardona
2004). These predispositions and conditions for seismic vulnerability typically exhibit physical, environmental, and socioeconomic aspects (Carreño et al.
2007; Barbat et al.
2009). We describe seismic vulnerability in this study as physical, socioeconomic, and environmental susceptibilities or conditions of limited capacity that facilitate harm in an earthquake event.
1.2 Vulnerability Assessment Method and Indices
Vulnerability assessment has become a method to communicate the disaster risk to the decision makers for risk reduction and adaptation (Birkmann
2006; Lindlay et al.
2011; Fekete
2019). A map of vulnerability supplemented with an estimate of the rate of people at risk is key to loss modeling and emergency and disaster risk management (Aubrecht et al.
2012). Vulnerability is a multidimensional construct that cannot be measured easily with a single variable (Cutter and Finch
2008). A Multi-criteria decision analysis (MCDA) with weight allocation is the most widely used technique to address urban vulnerability (Diaz-Sarachaga and Jato-Espino
2019) where a composite of multiple quantitative variables with an aggregative formula results in a single index score, called vulnerability index.
Several physical, environmental, and socioeconomic factors are considered responsible for producing the vulnerability of places (Carreño et al.
2007; Cutter and Finch
2008; Carreño et al.
2017). Although physical vulnerability that represents the fragility of the physical structures is crucial, comprehensive or holistic earthquake vulnerability covering overall predispositions is a far more encompassing concept (Barbat et al.
2009). Social vulnerability is described as socioeconomic status and characteristics that make people/societies susceptible to hazards (Cutter et al.
2003). Even though physical and social conditions may have diverse courses of change over space and time, they are considered to be inextricably linked together to produce high-risk conditions, where the former is indicative of the latter (Rashed and Weeks
2003). Socioeconomic conditions often force particular communities to live in hazard-prone areas and homes of poor structural condition (McEntire
2012) and are considered the causes of the physical vulnerability in many cases (Barbat et al.
2009). Built environment capacities, determined by the grain, density, and distribution of facilities such as parks, and so on, affects location-specific vulnerabilities (Wamsler et al.
2013). These inequalities across the space produce location-specific vulnerabilities communicated formerly as “hazard of places” (Cutter
1996).
A common interest of vulnerability assessment is representing the diversity of vulnerability across space. Congruently, mapping of earthquake vulnerability has become a trend in disaster risk reduction studies. The complex and multifaceted nature of the vulnerability concept is also reflected in a broad spectrum of spatial scales, for example, national, regional, local, and household levels (Diaz-Sarachaga and Jato-Espino
2019; Jaimes et al.
2022). Studies that assess intracity variation of vulnerabilities commonly use census units, for example, ward or neighborhood, or the census blocks, as mapping units. Studies that take smaller units as a basis, for example, buildings, are scarce and usually focus on the structural vulnerability (Zhai et al.
2019; Pavić et al.
2020), or they are based on standalone community surveys, rather than open census data (Birkmann
2006; Ebert et al.
2009). However, in urban environments of complexity, interactions across space generate spatial variations of characteristics regardless of the administrative units or census tracks that are for management purposes. Assigning an aggregate measure of characteristics of elements to units or groups, without any relation between them being demonstrated, may also lead to methodological error called “ecological fallacy” (Jones and Andrey
2007), which may cause drawing inaccurate conclusions about the vulnerability and the households/individuals that exhibit it. At the microscale, it becomes essential to represent spatial variation based on the smallest unit, that is, buildings.
In constructing a vulnerability index, the selection of indicators and the weighting are considered major challenges (Zhang et al.
2017; Ziarh et al.
2021). Although numerous studies have used equal weights (Zhang et al.
2017; Diaz-Sarachaga and Jato-Espino
2019), a common disposition is that indicators’ contribution to overall vulnerability is not equal, and weighing the indicators based on their relative importance is more likely to represent the real situation. There are broadly two approaches to weighing—the subjective and objective methods—that both have certain drawbacks. The subjective methods—the so-called knowledge-driven or expert-based methods—rely on the experts’ rankings or a consensus. The vast amount of studies that employ knowledge-driven methods such as AHP, TOPSIS, and so on, however, suffers from the subjectivity that introduces uncertainty and controversy into an index (Chen et al.
2010; Zhang et al.
2017; Rodcha et al.
2019). The objective methods—the so-called data-driven methods—attempt to overcome the subjectivity of the knowledge-based methods by determining weights based on the data. However, weights derived from data may not exactly reflect the actual weights of indicators and may deviate the results from the real situation (Zhang et al.
2021). To overcome the challenge of weighing, this study adopted the catastrophe progression (CP) method that uses relative importance of indicators, rather than direct use of hard to determine weights (Zhang et al.
2017).
Catastrophe progression, as a data-driven approach stemming from the study of dynamic systems, combines catastrophe theory with fuzzy mathematics (Cheng et al
1996) to develop a fuzzy membership function of system state (Gao et al.
2020). In CP, responses to changes in the internal values of each factor are intrinsically evaluated considering their ranking of importance that reduces subjectivity (Ahmed et al.
2015; Xue et al.
2022). Song et al. (
2020) and Mostafa (
2022) stated that the CP method has been demonstrated to have a unique advantage in dealing with uncertainty and is increasingly employed in holistic indices, particularly the vulnerability indices in the last decade (for example, Zhang et al.
2017; Ziarh et al.
2021; Zheng and Huang
2023). The CP method is also characterized by its perceptiveness to gradual changes in a system that may cause sudden shifts, resembling a catastrophe. Conventional data-driven methods that use calculus to reach a vulnerability index score, however, may fail to address multiple characteristics of an individual or neighborhood that interact and amplify each other to a fragile condition. We consider vulnerability as an example to this. Intrinsically, the CP method favors the relative importance of control variables where, ranking indicators regarding their importance is an issue to be addressed (Shen et al.
2020; Du et al.
2022). The CP executes hierarchical recursive calculations apt for a system tree. Therefore, rather than ranking indicators all at once, grouping indicators with shared context into categories and then ranking them is a compromise approach (Shen et al.
2020). The ranking is either knowledge-based or data-driven. By employing data-driven methods, for example, the mean square difference method (Jin and Zhang
2020), weights based on components share on total variance in principal component analysis (PCA) (Wood et al.
2010; Aksha et al.
2019), and entropy weights (Wu et al.
2022), subjectivity can be avoided. We adopted an extension of the CP method advancing entropy weights for objective ranking of the indicators.
The current study is motivated to objectively address the multidimensional nature of earthquake vulnerability. We acknowledge that earthquake vulnerability exhibits physical, socioeconomic, and environmental aspects and adopted a holistic approach that accommodates the below items as components of earthquake vulnerability index:
1.
Building vulnerability, as the structural fragility of the exposed residential building stock (construction system, period, plan irregularity, and so on);
2.
Socioeconomic vulnerability that produces susceptibilities for individuals and households (age, gender, household structure, income, and so on); and
3.
Vulnerability of the built environment that may reduce the coping capacity (building density, distance to assembly areas, and so on).
To conduct the analysis at a level of detail that can represent the vulnerability as a continuum across space and avoid ecological fallacy, vulnerability mapping at the microscale, taking buildings as the smallest unit of analysis, was adopted. Spatial clustering that advances communication by adding analytical abilities (Aksha et al.
2019) was employed. A classification was conducted on the vulnerability scores to reveal an estimate of the number of people that are highly vulnerable.
In the present study, indicators of earthquake vulnerability were chosen based on the relevance and availability of the data at the building and household scales. We particularly sought to employ data available or can be calculated using GIS tools at no cost.