1 Introduction

In the recent past, Australia has experienced several catastrophic hazard events, including bushfires, floods and severe storms. The frequency and intensity of such events is expected to increase in the future (Steffen 2009; Meehl et al. 2007). The impacts of recent disasters like the 2011 floods in Brisbane and the 2009 Victorian bushfires are significant, costing billions of dollars and hundreds of lives. Hazard-prone societies are always at risk for potential natural disasters. They can rarely be fully prevented, yet their losses can be minimized if the necessary preparedness and mitigation actions are taken before an event occurs and actions are undertaken to manage the hazard (Buckle 1998). Therefore, emergency managers, local governments and communities are under tremendous pressure to accomplish the task of vulnerability reduction. One way to accomplish this task is through measuring vulnerability (Birkmann 2006). Quantifying and visualizing vulnerability will help to provide an effective platform for decision makers to monitor and capture aspects of vulnerability (Kienberger et al. 2009).

In Australia, the majority of bushfire impacts occur at the urban–bush interface (UBI) (Buxton et al. 2011; Dixon 2005). Climate change scenarios predict that south-eastern Australia will experience increased temperatures, decreased precipitation and high winds, which will result in increased bushfire intensity, longer fire weather seasons and increased frequency of extreme or catastrophic fire risk days (Hennessy et al. 2005; Blue Mountains Bush Fire Coordinating Committee 2008). Population growth and urbanization in areas of high fire risk will increase the potential for loss. Bushfires at the UBI can threaten multiple dwellings very quickly and often overrun local resources (Lowe et al. 2009). Therefore, fires at the UBI pose the greatest challenge facing Australian fire agencies today. To utilize scarce response resources most efficiently, it is important to have prior knowledge about the level of vulnerability within communities at the UBI. Such information can also help to identify the most vulnerable populations to minimize the potential losses from future bushfire events by undertaking planning and preparation activities. Losses can be minimized if bushfire management activities are designed to target the most vulnerable people.

Vulnerability arises from the relationship between human actions and hazard impacts and has been extensively discussed since the 1970s (O’Keefe et al. 1976). The scientific community has put much effort into increasing our understanding of the concept of vulnerability and developing tools to assess the level of vulnerability within a population. Several conceptual frameworks have been developed to explain vulnerability to natural hazards. Widely recognized models include the risk-hazard (RH) model (Davidson 1997), the pressure and release (PAR) and access models (Blaikie et al. 1994; Wisner et al. 2004), the comprehensive vulnerability analysis (CVA) model (Turner et al. 2003), the vulnerability of place model (Cutter 1996; Cutter et al. 2000) and several other integrated models (UNISDR 2004; Dolan and Walker 2006; Füssel 2007).

When translating different conceptual frameworks into practical vulnerability analysis tools, researchers either have considered individual dimensions of vulnerability such as physical (Douglas 2007), biophysical (Cutter et al. 2000) or socio-economic (Clark et al. 1998; Cutter et al. 2003) or have taken a more holistic and integrated approach that addresses multiple dimensions of vulnerability (Preston et al. 2009; Dolan and Walker 2006). Socio-economic dimensions of vulnerability are a key element of many vulnerability frameworks (Füssel 2007; Hewitt 1997; UNISDR 2004; Wisner et al. 2004). The linkages between unsafe conditions and marginal socioeconomic conditions have been widely investigated in disaster management under the rubric of “Social Vulnerability” (Wei et al. 2004; Myers et al. 2008; Clark et al. 1998; Cutter et al. 2003; Wood et al. 2010; Vincent 2004). In this research, we focus on social dimensions of vulnerability to bushfires at the UBI.

The literature on social vulnerability and bushfires is not well developed compared with that of other hazards. In this research, we reviewed the literature on social vulnerability to other hazards to inform our work on bushfires. While several indicators of social vulnerability have been developed in Australia (Baum et al. 2008; Brunckhorst et al. 2011; Pink 2006), none of them have been applied in the context of bushfires. The lack of work on social vulnerability and bushfires was addressed by the Victorian Bushfires Royal Commission (2009), which highlighted the importance for effective emergency response of having demographic information on vulnerable populations. Such information can guide emergency managers to recognize the specific needs of vulnerable people such as those who might need evacuation assistance or separate consideration, particularly on code red (high fire risk) days (Victorian Bushfires Royal Commission 2009).

Constructing a social vulnerability index is a common approach to assessing social vulnerability to hazards. Such an index helps to identify where socially vulnerable groups are located (Fekete 2009). Vulnerability indices can help planners to prioritize locations of populations with high levels of social vulnerability when they make emergency and risk management decisions (Chakraborty et al. 2005) as well as to see variations in social vulnerability over time and across space (Cutter and Finch 2008).

We tested whether a standardized social vulnerability index could be developed using a pre-existing set of indicators, by creating an exploratory principle component analysis model using Australian Bureau of Statistics (ABS) 2006 census data at the Census Collection District (CCD) level. Two communities with considerable biophysical exposure to bushfire at the UBI were investigated: the Blue Mountains council area to the west of Sydney and the Ku-ring-gai city council area in the north of Sydney, Australia. We identified variables contributing to social vulnerability and used the component scores to develop a social vulnerability index. Finally, we mapped the social vulnerability index at the CCD level.

2 Social vulnerability

Social vulnerability in particular refers to demographic and socioeconomic characteristics of a population that influence its capacity to prepare for, respond to and recover from hazards and disasters (Cutter et al. 2009). This difference in capacity may be one reason why communities with similar levels of exposure may experience different impacts from a particular hazard event (Finch et al. 2010; Cutter et al. 2009). Over the last decade, the importance of social vulnerability has been increasingly recognized and it is now considered to be an important part of understanding disaster risk. It can be considered a by-product of social inequalities, which increase an individual or population’s level of susceptibility to a particular hazard by limiting access to opportunities to prepare, respond and recover (Cutter et al. 2003; Yeletaysi et al. 2009). Some people struggle to cope with their everyday life due to social conditions such as low income, occupation, etc. For such populations, disaster preparedness is not the priority.

Social vulnerability can affect all phases of the disaster cycle. However, its effects are most visible in the response and recovery phases. For example, in Australia, people in non-English speaking households may be slower to receive and respond to early warning and evacuation orders (Goudie 2008). During the recovery phase, people who can afford to insure their properties and who have better savings will recover faster than people who are financially deprived. Socially vulnerable groups are often the least prepared for a hazard as their ability to prepare is hampered by their limited financial resources and/or they may experience physically constrained access to preparedness, response and recovery supplies (Tapsell et al. 2010). For example, low-income groups are less likely to have insurance, less likely to receive and believe disaster management messages and less likely to act on orders that will help them mitigate their experience of a disaster (Zahran et al. 2008).

2.1 Social vulnerability assessment

Indices and indicators of disaster risk and vulnerability at national and sub-national scales are the most commonly used tools for identifying and measuring vulnerability to hazards (Birkmann 2006). One common practice is constructing a Social Vulnerability Index (SoVI), which is an indicator-based approach that relies on socioeconomic and demographic data to model social vulnerability. SoVI methods reduce a large number of variables into a few statistically significant dimensions that describe different aspects of social vulnerability. Then, a model (i.e., an index) is developed to quantify social vulnerability based on these dimensions (Schmidtlein et al. 2008; Adger et al. 2004). In general terms, an index is defined as a composite representation of several numerical measurements within one single value (Simpson and Katirai 2006). The goal in building an index is to reduce the complexity of information without diminishing the information’s capacity to shed light on the problem (Bowen and Riley 2003 in Adger et al. 2004). Use of indices has gained wide popularity because they are easier for the public to interpret than identifying common trends across many individual variables (Saltelli 2007). They are also useful for benchmarking a given condition, and for identifying trends across space and through time (Nardo et al. 2005).

However, the effective use of indices presents some challenges. Composite indices may send misleading policy messages if they are poorly constructed or misinterpreted (Nardo et al. 2005; Saisana et al. 2005; Saltelli 2007). Therefore, when constructing an index, the selection of indicators, choice of model, the weighting of indicators, and treatment of missing values need to be addressed carefully. Choosing indicators requires extensive review of prior research, and the construction process should be transparent and based upon robust statistical and conceptual principles (Nardo et al. 2005; Saisana et al. 2005; Saltelli 2007). With respect to interpretation, index values only indicate relative differences. Therefore, no significance should be attached to the absolute values of the index, and index values can only be used to compare different groups, entities, or geographic areas in order to perform ranking. Ranking is useful when targeting and in identifying priorities (Eakin and Luers 2006). Despite these challenges, vulnerability indices are widely applied in the context of disaster research (Clark et al. 1998; Cutter et al. 2003; Cutter and Finch 2008; Holand et al. 2009; Wood et al. 2010).

The concept of a social vulnerability index was introduced by Cutter et al. (1997), who presented a simple method of developing a single score to measure social vulnerability at the US census block level using identified social vulnerability indicators. In 2000, Cutter et al. operationalized this approach as a component of their “hazard of place” model. In their method, eight indicators that describe characteristics such as population size and structure, differential access to resources and physical weaknesses, wealth or poverty, and level of physical or structural vulnerability were selected to quantify social vulnerability based on prior knowledge of socioeconomic and demographic characteristics and their relationships with hazard impact. Standardized values of those indicators were summed to develop an aggregate measure of social vulnerability. Since then, their method has been used as an illustrative tool to capture the societal conditions of populations living within a given geographic unit (Chakraborty et al. 2005; Wu et al. 2002; Cutter et al. 2000).

Clark et al. (1998) employed a more complex method of analyzing social vulnerability based on factor analysis. Vulnerability factors (poverty, transience, disabilities, immigrants, young families) were identified based on census data. In 2003, Cutter et al. further developed the social vulnerability index approach using principle component analysis. In this study, they identified components that describe different dimensions of social vulnerability. This method has been widely accepted and utilized in many studies (Fekete 2009; Rygel et al. 2006; Mendes 2009; Wood et al. 2010).

Despite being widely used, there is not yet agreement about how SoVI components should be aggregated to yield a composite index. The most common method of aggregation is a non-weighted additive model in which all components contribute equally to the final index (Cutter et al. 2003; Fekete 2009; Mendes 2009; Myers et al. 2008). Other researchers have used a weighting scheme that is based on the percent of variance explained by each component (Cox et al. 2007; Wood et al. 2010). Schmidtlein et al. (2008) used both non-weighted additive and weighted sum methods to analyze the sensitivity of their SoVI. They found substantially different results using the weighted sums and equal weights approaches. A third strategy assigns weights based on other criteria. For example, Rygel et al. (2006) used a Pareto ranking of factors, while Haki et al. (2004) assigned weights based on a pairwise comparison technique carried out by experts from different disciplines such as employment, housing, education, and household demography. Because of the different results that these different methods produce, it is important to try to verify and validate the results of a SoVI analysis, insofar as possible.

Fekete (2009) validated the SoVI approach using an independent second dataset describing the sociodemographic characteristics of a population exposed to river floods in Germany. He first created a census-derived SoVI that predicted who would be socially vulnerable to river flooding. He then used a survey that identified individuals who were vulnerable to a real flood event to test the predictive power of his model. To do this, he used a logistic regression of the independent variables in his vulnerability index to predict the probability of whether an individual with particular sociodemographic characteristics would be forced to leave their home or seek emergency shelter. The variables that were significant in this analysis were retained, and he then repeated the factor analysis that generated his original SoVI with these retained variables and compared the two indices. His results revealed that the regression-captured variables yielded the same factors that his original SoVI index contained. He concluded that these results demonstrated the validity of the original index. Fekete’s approach, however, is difficult to replicate in the absence of a hazard event and data on who was actually vulnerable during that event.

Vulnerability is a product of processes occurring at different spatiotemporal scales. Measuring vulnerability at different scales helps to understand cross-scalar dynamics in vulnerability, from global to local or vice versa (Chen et al. 2003). Selecting an appropriate scale is crucial and often challenging as the relevance of a specific variable may change with scale and degree of data aggregation (Turner et al. 2003; Fekete et al. 2010; Eakin and Luers 2006). The scale of analysis plays an important role in identifying vulnerability of people and places as spatial distribution of vulnerability varies with the scale. While sub-national scales provide insight into regional processes and patterns, finer scales are beneficial for better understanding and capturing the root causes of vulnerabilities (Fekete et al. 2010). In the context of disaster management, vulnerability assessments are often conducted at scales that coincide with the geographic levels at which disaster management decisions are taken (Cutter et al. 2008; Eakin and Luers 2006).The selection of appropriate scale is often influenced by data availability and quality, construction and presentation of outputs, and the extent of the study area.

2.2 Social vulnerability indicators

Assessments of social vulnerability have frequently utilized different indicators and proxies that describe the socioeconomic and demographic characteristics of individuals and communities and that influence the level of social vulnerability in a particular geographic location. Indicators are selected based on either theoretical understanding of their relationships to vulnerability or based on statistical relationships (Adger et al. 2004). Such indicators can be selected from the wealth of information about the population normally readily available through a national census. Selection of indicators is problematic due to complex interactions between variables. Therefore, social vulnerability indices include varying degrees of detail depending upon the context and the information, data and resources available.

Widely accepted characteristics that influence social vulnerability include income, gender, race/ethnicity, age and level of education (Cutter et al. 2003; Rygel et al. 2006; Cutter et al. 2000). In some studies, characteristics of the built environment such as building density and building age have also been incorporated (Mendes 2009; Holand et al. 2009). Studies that have used these variables have defined social vulnerability as a function of socioeconomic, demographic and built environment characteristics. The aim of this study is to capture the socio-demographic dimensions of the people who are living at the UBI. Therefore, we only focus on the socio-demographic attributes of communities and exclude built environment variables.

3 Study area

The urban–bush interface (UBI) is a zone where urban and peri-urban development and activities overlap with natural or semi-natural bushland vegetation (Lein and Stump 2009; Collins 2005). Outside of Australia, this area is often described as the wildland–urban interface (WUI). In the context of bushfires, the UBI is “any area where structures (whether residential, industrial, recreational, or agricultural) are located adjacent to or among combustible wild land fuels” (Cottrell 2005, p. 110). Due to increasing human population levels and demands for both affordable housing and lifestyle blocks, urban development is expanding deeper into bushland areas that are fire prone. This expansion has put many communities at increased bushfire risk. In particular, in recent years, bushfire risk at the UBI has increased because of the growing popularity of rural residential living and the preservation of inherently flammable natural areas within UBI development (Granger and Hayne 2001). For this study, we selected the UBIs within the Ku-ring-gai and Blue Mountains Local Government council areas because both locations are at high biophysical risk of bushfire, yet they have differing sociodemographic characteristics.

3.1 Blue Mountains LGA

The Blue Mountains Local Government Area (LGA) (Fig. 1) is a major tourist destination in Australia because of its natural beauty and location at the western fringe of the Sydney Metropolitan Area. It is nestled within a larger World Heritage Area proclaimed in 2000, due to its outstanding level of unique natural values. The Blue Mountains LGA covers 1,433 km2, of which 70 % comprises the Blue Mountains National Park. The LGA has a population of approximately 76,719 within 26 suburbs. Blue Mountains suburbs can be classified as belonging to the Lower Blue Mountains (Lapstone, Glenbrook, Blaxland, Mount Riverview, Warrimoo, Sun Valley, Valley Heights, Springwood, Winmalee, Yellow Rock and Faulconbridge), Mid-Mountains (Linden, Woodford, Hazelbrook, Lawson and Bullaburra) and Upper Blue Mountains (Wentworth Falls, Leura, Katoomba, Medlow Bath, Blackheath, Mount Victoria and Bell). However, the area’s population is largely concentrated in Springwood and Katoomba, the major town centers. The major industry in the Blue Mountains is tourism, which contributes significantly to the economy, industrial and cultural settings of the area.

Fig. 1
figure 1

Map showing the study areas and the greater Sydney region

Full size image

A high proportion of the LGA’s permanent residents was born in Australia. Residents’ average taxable incomes are low compared with the broader Sydney metropolitan area. Moreover, the income differential between area residents and those of the Sydney metropolitan area is increasing (City of Blue Mountains 2009). There is a gradient of disadvantage across the Blue Mountains LGA (SEIFA scores ranging from 810 to 1,152), with people living in the Lower Blue Mountains tending to be more advantaged in terms of household income, educational achievement and employment opportunities than those in the Upper Blue Mountains (Blue Mountains City Council 2006). This relationship is likely due to the Lower Blue Mountains’ greater access to the resources of the broader Sydney Metropolitan Region.

The Blue Mountains has numerous settlements situated in highly bushfire-prone areas, with most communities located on the exposed UBI in close proximity to bushland. Almost three-quarters of residential buildings (73 %) in the LGA are in high bushfire risk areas (Chen 2005). Therefore, based on the percent of addresses located within bushfire-prone land, the Blue Mountains LGA ranks as the most bushfire-exposed LGA out of the 61 LGAs in the Greater Sydney Metropolitan Region (Chen 2005).

The bushfire season in the Blue Mountains runs from September to February. Favorable weather conditions associated with westerly or northwesterly winds, high temperatures and low humidity can lead to extreme fire behavior with potentially catastrophic consequences. This poses threats to life and property, which creates challenges in emergency management and preparedness for bushfires. In addition, the area’s topography and limited road access makes the evacuation process extremely difficult in the event of bushfires (Burton and Laurie 2009). Because of the presence of tourists, on any given day there may also be large numbers of people who have little knowledge of bushfire risk present in the area.

The Blue Mountains area has on average 28 bushfires per year (Blue Mountains Bush Fire Coordinating Committee 2008). Of these, seven can be categorized as severe fires that run longer than 24 h. The main sources of ignition in the Blue Mountains are human activities and lightning. Human-caused ignitions include deliberately lit fires and improperly supervised campfires. Ignition from lightning occurs due to thunderstorm activity.

3.2 Ku-ring-gai LGA

The Ku-ring-gai LGA is located in the northwest suburbs of Sydney (Fig. 1) and lies 15 km from Sydney’s CBD. It much more urbanized than the Blue Mountains LGA. The total council area spans 86 km2, of which 32 km2 are covered by bushland. The LGA is surrounded by national parks on three sides; Ku-ring-gai National Park to the east, Lane Cove National Park to the west, and Garigal National Park to the southwest.

The LGA has an estimated population of 108,135 within 17 suburbs (Australian Bureau of Statistics 2009). A majority of the residents of Ku-ring-gai are either Australian-born or from an English speaking background. They are well-established in the area, with most having lived there for more than 10 years. Most residents own their houses. In Ku-ring-gai, a large proportion of residents are married and live in family households. Residents’ levels of education are high, but the level of full time employment is relatively low because a high proportion of the area’s residents are retired.

The Ku-ring-gai LGA is well-known for the quality of its primary and secondary school facilities. Major hospitals and nursing homes are also found in the area. In general, Ku-ring-gai’s socioeconomic indicators such as level of income, education and occupation are higher than the Sydney average (Ku-ring-gai Council 2005, 2006; Deodhar 2004).

Ku-ring-gai is ranked as the third most vulnerable LGA to bushfire out of 61 LGAs in the Greater Sydney Metropolitan Region, based on the percent of addresses located within bushfire-prone land (Chen 2005). Bushfire season in Ku-ring-gai runs from October to March. During these months, high daytime temperatures and strong northwest winds typically prevail. Low humidity and dry vegetation along with high fuel loads and prevailing weather conditions influence the bushfire behavior in the area. There is a high probability of bushfire occurrence in the months of December and January (Hornsby/Ku-ring-gai Bushfire Management Committee 2008).

Ku-ring-gai has experienced severe bushfires that have caused significant damage to property and threatened lives. Ku-ring-gai and Hornsby Councils work together on bushfire management as the national parks border both areas. On average, 40 bushfire incidents are reported per year across both areas. Every 7–10 years, a major bushfire event is expected. The main sources of ignition in Ku-ring-gai are human activities or lightning. Human caused ignitions include deliberate ignitions, campfires and debris burning (Hornsby/Ku-ring-gai Bushfire Management Committee 2008).

4 Methodology

This study relies on the method developed by Cutter et al. (2003). To describe social vulnerability at the UBI, we developed a composite index based on 2006 Australian Bureau of Statistics (ABS) Census data, the most recent data available. This index provides a relative measure of socioeconomic and demographic conditions at the CCD level. It also provides profiles of who lives where to further help to understand the types of residents living in the UBI and their social vulnerability.

We initially selected 51 variables from the 2006 Australian Census data to capture indicators of socioeconomic status, gender, age and disabilities, family structure, housing, education, migrants and cultural values (Table 1). Our selection of variables was guided by those previously reported in SoVI methods in the literature (Cutter et al. 2000, 2003; Fekete 2009; Flanagan et al. 2011; Rygel et al. 2006; Wu et al. 2002). The spatial scale of this study is the CCD level, because that is the smallest unit available with complete data. Often community-level bushfire management initiatives are conducted at the street level and the CCD scale is best suited for such initiatives. We obtained data for the 239 CCDs that cover the UBIs in the Ku-ring-gai and Blue Mountains LGAs. Several CCDs contained no population. These CCDs are sites that have been recently developed and that contained no residents in 2006. Therefore, we removed these CCDs from the analysis and labeled them as containing “no data”.

Table 1 Social vulnerability indicators and variables used in the analysis
Full size table

Our SoVI employed Principle Component Analysis (PCA). PCA is a multivariate statistical technique that is frequently used as a data reduction method. It decomposes an original set of variables into a smaller number of linear variates (Field 2009). PCA works well when the distribution of variables varies across cases, or in this instance CCDs. The result of PCA is a set of uncorrelated components that represent a linear weighted combination of the initial set of variables (Vyas and Kumaranayake 2006). We then used the resulting components to create a social vulnerability index.

We tested the original variables for multicollinearity as PCA does not produce robust results when input variables are perfectly correlated or completely uncorrelated (Field 2009), retaining 21 variables from Ku-ring-gai and 25 variables from the Blue Mountains for the analysis. We created density, per capita or percent functions for all remaining variables. This enabled us to compare CCD units of different sizes. These transformed variables were then standardized so that variables could be compared with each other. Because we performed the PCA separately for each LGA, our results can be compared within one LGA but not between LGAs. We deliberately made this choice to investigate whether the same variables would be important for describing social vulnerability in different parts of the Sydney basin.

We performed the PCA using SPSS (SPSS 18.0) using varimax rotation to find the best possible fit to the dataset in order to calculate the highest variance for each component. We applied the Kaiser criterion (eigenvalues >1) for component selection. We carried out the stepwise exclusion approach and repeated this process until the variables and components of the model were stable and statistically robust. To check the robustness of the model, we used the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity and communalities. Finally, we adjusted the sign of the components according to their directional influence based on literature reports (Cutter et al. 2003). We considered components that increased vulnerability to be positive and components that decreased vulnerability to be negative (Wood et al. 2010). We did not have any variables that loaded both positively and negatively on a component, so we did not apply any absolute value transformations as did Cutter et al. (2003).

Finally, we developed a composite index by adding weighted component scores. We weighted each component score based on the percent of variance explained by the component divided by the total variance explained by all extracted components, in order to produce weights that ranged in value from 0 to 1. We classified the final SoVI using the standard deviation from the mean method, which provides a relative measure of deviation from the mean for each CCD (Wood et al. 2010). Finally, we mapped the results at the CCD level to illustrate their spatial variability. In order to compare the level of social vulnerability in the area, we classified CCDs with a SoVI higher and lower than 1.5 positive or negative standard deviations from the mean as high and low, respectively. We also mapped individual component scores using the quantile classification method to facilitate comparisons between components. Individual component score maps illustrate the spatial variability of each component score among CCDs. They also highlight the dimension(s) of social vulnerability that are most important in particular locations.

Vulnerability models can be validated by comparing their predictions with an independent dataset that includes measures of specific post-event outcomes (Fekete 2009). Such data sets are not often available and therefore validation of SoVIs is difficult. Hence, few authors have validated their SoVI models (Clark et al. 1998; Cutter et al. 2003; Mendes 2009; Wood et al. 2010). No household level post-event data set was available for our study areas. Therefore, validating the result empirically was not possible. Nevertheless, we used expert interviews with council employees, community fire volunteers and community members and observations on field trips to check that our SoVIs’ results were consistent with local knowledge about social vulnerability. This verification process led us to conclude that our SoVI components broadly captured the level of social vulnerability in the area.

5 Results

The PCA generated six components for both the Ku-ring-gai and Blue Mountains LGAs. In both locations, the analyses explained more than 70 % of the total cumulative variance in social vulnerability. One component was common across areas: Immigrants, which explains 12 % of the variance in Ku-ring-gai and 10 % in the Blue Mountains (Table 2).

Table 2 Social vulnerability index comparison
Full size table

5.1 Dimensions of social vulnerability in Ku-ring-gai

Results of the PCA (Table 3) identified six components of social vulnerability. A KMO of 0.754 indicates that the variables were suitable for analysis and the extracted commonalities were above 0.5, which indicates that the extracted components represent the variables well (Fekete 2009).

Table 3 Social vulnerability and its dimensions—Ku-ring-gai LGA
Full size table

The first component, Special Needs, Elderly and Gender, explained 25.79 % of the total variance and included variables such as the percent living alone (+), percent needing special assistance (+), percent of housing units with no internet connection (+), percent over 65 (+), percent of dwellings with no vehicle (+), percent female (+) and percent who did unpaid domestic work less than 5 h per week (−). We indicate the direction of loading for each variable in brackets. People who live alone, the elderly, people who need special assistance, and females may be more likely to need assistance from others in the event of a disaster and are therefore more vulnerable than those who have family connections, are young, do not need special assistance and males. Access to information and ability to act upon that information during response and recovery is an important component that determines the level of social vulnerability. Households with limited independent mobility are also vulnerable during a time of evacuation (Granger 2003; Chakraborty et al. 2005). Households with no internet connection or vehicle may have difficulties during a disaster event, increasing their social vulnerability. Some women may be more vulnerable to disasters. For example, a majority of single parents are women, and women from some cultural backgrounds, including those found among migrants to Australia’s cities, may be disempowered (Dwyer et al. 2004). Unpaid housework is also important because it is possible that well-maintained properties are somewhat better prepared (and hence less vulnerable) than more poorly maintained ones (Poudyal et al. 2012).

The second component, Tenancy and Housing, accounts for 12.54 % of the total variance. It includes three variables: percent of housing units that are rented (+), percent who moved house in the past year (+) and percent of dwellings that are cooperative, community or church housing units (+). Ownership and quality of housing is an important factor in assessing social vulnerability. In general, rental houses are not as well-maintained as owner-occupied houses and renters are less likely to take actions to protect property from a disaster (Cutter et al. 2003; Kuhlicke et al. 2011). Often, low-income levels limit the range of dwelling type available for renters’ families (Baum et al. 2008). People who have recently moved may be new to the area, and their level of awareness of local risk conditions is likely to be low. Levels of available individual financial resources are also typically lower for both renters and transient populations.

The third component, Immigrants, is comprised of the percent of the population that speaks a language other than English at home (+), the percent who do not speak English well or at all (+) and the percent of migrants from overseas (+). This component represents 12.28 % of the total variance. Immigrants are vulnerable to natural hazards due to their different cultural backgrounds and racial and ethnic inequalities that often lead to discrimination against them (Wood et al. 2010). Often immigrants lack experience with local disasters, and their awareness of risk is also low because they may have limited English proficiency.

The fourth component, Wealth, captures three variables and explains 10.04 % of the total variance. The variables include median individual income (+), median household income (+) and percent of people with incomes of $1–$249 per week (−). People with higher levels of income have better capacities to prepare for and recover from disasters. People with low levels of income often do not have any disaster recovery plans such as insurance and may have few savings or other assets. Therefore, low-income households are often disproportionately impacted by a disaster (Cox et al. 2007; Flanagan et al. 2011; Wood et al. 2010).

The fifth component is Occupation, accounting for 7.92 % of the variance. It is comprised of the percent who are employed as managers and professionals (−) and median rent per week (+). Occupation contributes to social vulnerability (Kuhlicke et al. 2011; Holand et al. 2011). People employed as managers and professionals often have higher incomes that influence their ability to prepare and recover from disasters.

The final component is Visitors and Low Education; it explains 7.87 % of the variance. It includes two variables, namely percent of visitors (+), and percent of people over 20 years of age who have a year 10 education or less (+). Visitors’ lack of familiarity with the local environment and emergency response planning, along with their relative isolation from the general community, makes them more vulnerable during disasters (Granger 2003; King and MacGregor 2000). Education is typically associated with an individual’s level of income and poverty. Moreover, a low level of education may also negatively influence a person’s knowledge and awareness of bushfire management practices.

5.2 Dimensions of social vulnerability in the Blue Mountains

We used the same initial set of 51 variables for the Blue Mountains LGA analysis; we retained 26 variables for the PCA. These variables represent six components, which describe 73.04 % of the total variance in the data (Table 4). A KMO of 0.829 indicates that the variables were suitable for analysis and the extracted commonalities were above 0.5.

Table 4 Social vulnerability and its dimensions—Blue Mountains LGA
Full size table

The first component, Tenancy and Household Wealth, is comprised of ten variables and explains 27.57 % of the variance. Percent of rented housing units (+) is the dominant variable. Other variables that contribute are percent of dwellings with no vehicle (+), percent living alone (+), percent of housing units with no internet connection (+), percent of one parent families with child under 15 (+), percent who moved house within the past year (+), percent of visitors (+), median household income per week (−), median rent per week (−) and median housing loan repayment per month (−). Negatively loading variables explain the economic circumstances that contribute to a household’s level of preparedness and recovery. Household resources such as income, vehicles and an internet connection are also very important during the emergency response phase because they support an effective response.

The second component contains four variables, which generally relate to Age and Special Needs. It explains 12.50 % of the variance. It is comprised of median age of persons (+), percent of the population aged over 65 (+), percent who need special assistance (+) and percent of children under 5 (−). The elderly and people who need special assistance are considered to be vulnerable to disasters (King and MacGregor 2000). This is due to their low levels of mobility and high levels of isolation. Children from the youngest age group (below 5 years) are one of the most vulnerable groups in a disaster event. They are heavily dependent on others because they cannot protect themselves (Flanagan et al. 2011).

The third component represents the Immigrant population, accounting for 10.33 % of the variance. It consists of four variables; percent speaking a language other than English at home (+), percent of migrants (+), percent of people with negative income (+) and percent of non-Australian citizens (+).

The fourth component describes Low Education. It explains 9.41 % of the variance and includes three variables; percent over 20 years who have a year 10 education or below (+), percent of people who provided unpaid assistance (+) and percent of volunteers (+).

The fifth component is Gender and Domestic Work, which explains 7.63 % of the variance and includes two variables: percent of females (+) and percent of people who did unpaid domestic work less than 5 h per week (−). The literature suggests that women are more vulnerable than men during disasters due to family care responsibilities and sector-specific employment such as personal services (Rygel et al. 2006; Tapsell et al. 2010).

The final component represents Occupation and Personal Wealth. It explains 5.58 % of the variance. The variables include percent employed as managers and professionals (+), percent of people with incomes of $1–$249 per week (−) and percent employed in service occupations (+). Occupation is an important factor that influences incomes. Low wage occupations, including many service occupations, are a group that might face slower recovery after a disaster (Cutter et al. 2003; Wood et al. 2010; Poudyal et al. 2012).

5.3 Social vulnerability index and mapping

With the increased use of geographic information systems and mapping techniques, visualizing the distribution of vulnerability has become a central tool for communicating the results of vulnerability research to decision makers and the general public (Eakin and Luers 2006; King 2001). They are being widely used to capture spatial and temporal changes in social vulnerability (Cutter et al. 2003; Cutter and Finch 2008; Fekete 2009; Schmidtlein et al. 2008). Preston et al. (2011) have identified twofold benefits of vulnerability mapping; motivating policy responses and social learning.

For any disaster management activity, a relative measure of vulnerability gives useful insights into potential hotspots of high vulnerability. A SoVI provides a relative measure of social vulnerability across CCDs. A single composite index was calculated for each CCD using component scores and their variances. We believed that not all components contribute equally to social vulnerability. Therefore, we weighted each component by the percent of variance it explained divided by the total variance explained by all components. CCDs with high index values represent high vulnerability whereas lower index values represent low vulnerability. In order to identify the spatial patterns of social vulnerability, we mapped the resulting SoVI scores at the CCD level (Fig. 2). CCDs were classified using a five category mean-standard deviation classification method, with classes ranging from below −1.5 SD to more than +1.5 SD. Then, we mapped each (unweighted) individual component to illustrate the spatial heterogeneity among the CCDs, using a quantile classification method to facilitate comparisons between components (Figs. 3 and 4).

Fig. 2
figure 2

Social vulnerability by CCD at the UBI based on SoVI

Full size image
Fig. 3
figure 3figure 3

Maps showing distribution of individual component scores by CCD in the Ku-ring-gai LGA

Full size image
Fig. 4
figure 4figure 4

Maps showing distribution of individual component scores by CCD in the Blue Mountains LGA

Full size image

In both LGAs, most CCDs show moderate levels of social vulnerability. In the Kur-ring-gai LGA, eight CCDs can be classified as highly vulnerable because their index values are higher than +1.5 SD. Social vulnerability is highly associated with the component Special Needs, Elderly and Gender in the communities in the area’s northern and eastern suburbs, such as North Turramurra and St Ives. In the southern part of the LGA, the level of social vulnerability is more closely related to the Tenancy and Housing component. The high vulnerability CCDs that are in close spatial proximity to the major highway in the area (it runs through the non-UBI region in the middle of the map) include several CCDs with high numbers of immigrants (Component 3, Fig. 3). The value of mapping individual component scores to aid in the understanding of the drivers of vulnerability in particular locations is well-illustrated by the high vulnerability CCD near Fox Valley, whose vulnerability derives from high values on components 3–6 rather than components 1 and 2.

In the Blue Mountains LGA, the results show that social vulnerability to bushfire is significantly higher in the suburbs of the Upper Mountains than in those of the Lower Mountains, particularly in the areas in and around Katoomba. A total of 13 CCDs fall into the highest vulnerability category. Eleven of these highly vulnerable CCDs are in the Upper Blue Mountains and two are in Springwood, which is located in the Lower Blue Mountains. The highest levels of social vulnerability are driven primarily by the Tenancy and Economy, Age and Special Needs components (Fig. 4). Immigrants is also an important component in the Upper Mountains. The Occupation and Personal Wealth component shows highly vulnerable CCDs largely concentrated around the Katoomba, Springwood and Blaxland town centers. Low education is evenly distributed across the study area. Most of the CCDs in the Mid-Mountains area show average or low levels of vulnerability for all the components except for Gender and Domestic Work.

6 Discussion

When considering the implementation of any mitigation, preparedness, response and recovery measure, it is important to address the particular social dimensions that make communities more vulnerable. Therefore, understanding the social dimensions of vulnerability in a community is a vital step in any efficient planning process. A SoVI summarizes the complexity of social dimensions. It is multidimensional in nature and explains the sensitivity of a population to natural hazards and its ability to respond to and recover from the impacts of hazards (Cutter and Finch 2008). Although a SoVI provides a relative measure of vulnerability, the individual component scores can be also be used to understand different dimensions that make a particular area socially vulnerable (Rygel et al. 2006). Mapping each component illustrates the spatial variability of that component among CCDs and can help emergency managers to plan and prioritize suitable strategies that target specific problems and specific populations at specific places.

SoVI is a relative measure. It only provides information on inequality between CCDs, and in our study, between CCDs within one LGA. SoVI does not provide information on absolute levels of socioeconomic and demographic characteristics within a CCD. Moreover, PCA is sensitive to variable selection (Fekete 2009). The choice of variables can have an impact on final index values. Therefore, special consideration is required during variable selection process.

In this study, we replicated our analysis for both the Ku-ring-gai and Blue Mountains LGAs. The components of social vulnerability and the contribution of each component to vulnerability levels were not exactly the same in both areas, an important point for local policy makers. Each community has its own social dimensions and variables that explain who is most vulnerable (Schmidtlein et al. 2008). Our study identified several components that are broadly consistent with other social vulnerability studies (e.g., Cutter et al. 2003. Rygel et al. 2006).

Our results reveal differences in the drivers of social vulnerability across space. The primary component that contributes to the overall level of social vulnerability was different among different CCDs. Social vulnerability maps, particularly those of the index’s components, provide deeper insights into what is important in shaping a particular community’s social vulnerability, which will help fire managers to identify and prioritize specific fire management programmes to meet the specific needs of particular vulnerable communities.

For effective wildfire mitigation and adaptation, fire management authorities should focus on places where socially marginal populations intersect with higher wildfire risk because these populations have the added vulnerabilities of lower capacity to prepare for hazard events (Poudyal et al. 2012). In NSW, several community-level bushfire management initiatives, such as Community Fire Units (CFU) and Street FireWise community education programmes, are being conducted. However, currently these units are not necessarily being established in places where they are most needed: those with the most vulnerable populations. Rather, they are located in areas of relative affluence where residents are often already among the most prepared (Greg Buckley 2010, personal communication).

While local response organizations (e.g., police, fire services) are typically aware of where socially disadvantaged populations can be found, social disadvantage is not the only correlate of vulnerability. Furthermore, although a standard index of socioeconomic disadvantage (SEIFA) is available in Australia, it excludes many variables that are very relevant to social vulnerability, such as people needing special assistance during fires and recent migrants. Therefore, the use of SEIFA on its own is insufficient to meet the needs of either a disaster planner or an emergency manager (King 2001). Moreover, a SoVI index provides a ranking of comparative levels of vulnerability that can be used to prioritize interventions when scarce resources do not allow for intervention in all vulnerable areas.

Most social vulnerability assessments have been conducted using census data and census geographies. In this research, we derived variables from the 2006 Australian Census. However, census data only provide some information about the communities, and the final index is influenced by available data. King and MacGregor (2000) and King (2001) have identified some of the limitations of this approach. For example, as administrative units, census geographies may not represent actual communities. Actual communities may cross the boundaries of census geographies. Residents’ perceptions, norms, values and cultural attitudes as well as the complexity of social interactions within communities may not be represented by census data.

Available census data also may not represent the current sociodemographics of an area. The Australian Census is undertaken in winter, when the characteristics of the population in a given location may not be the same as during the summer and school holidays. The census also measures nighttime population, which differs in important respects from daytime populations. On weekdays, most of the nighttime population may not be at home, and only retired persons, the elderly and young children (and their carers) are at home. These population characteristics are hard to capture with census data. Nevertheless, for many communities, the census remains the best available source of population data because the data are consistent, repeatable and readily available. Yet, these limitations mean that a critical examination of the map, its underlying data and assumptions is required before decisions are made based upon the map (Preston et al. 2011).

The assessment of social vulnerability using a SoVI is not an end product. Knowledge about the geographic and sociodemographic context is required to link the results of a SoVI analysis with the real world. When possible, SoVIs need to be validated to verify the relationship between social dimensions and hazard impacts using an independent secondary data set (Fekete 2009).

In this research, we used local knowledge and data on exposure to recent bushfire events in other locations in Australia to verify and interpret our results. We could not directly validate the SoVI generated in this research since a secondary dataset on socio-demographic characteristics of a bushfire-affected population for the study area is not available. Indeed, Adger et al. (2004) have noted the difficulty in general of validating vulnerability indicators because hazard events in a given location are relatively rare. This is exacerbated by the fact that the characteristics of populations in a given location change over time, so validation requires relatively recent data on a population’s experiences of a hazard to identify who was vulnerable. However, the lack of a validation data set does not mean that some method of assessing the levels of vulnerability within a community is not needed for hazard management. In the absence of a validation data set, we can get some indication of whether our results are reasonable by examining populations who were vulnerable to recent bushfire events in other parts of Australia.

Recent studies on bushfire impacts in Australia have shown that social vulnerability and related processes that place specific segments of human populations at risk can create a disaster when a bushfire strikes. Reviews of fatalities in the Black Saturday bushfires in Victoria show that many of those who died in the event (i.e., those who were vulnerable; over 44 %) were disabled, elderly or very young (i.e., children) (Handmer et al. 2010). In both of our LGAs, these variables were included in either the first or second principle component (i.e., those that explain a large proportion of variation in the data). Unfortunately, the Black Saturday fatality report did not detail any other sociodemographic characteristics of the affected population that we could use for validation. Our lack of access to a validation data set is a limitation of this study.

7 Conclusions

Prevailing policies and disaster management frameworks should address social inequalities and their impacts on disaster preparedness, and social vulnerability needs to be better integrated into current hazard management planning. Social vulnerability information can be used at any stage of the disaster management cycle. At the mitigation and preparedness phase, SoVI can be used as a proactive tool to plan community-based disaster mitigation activities that support more vulnerable areas in order to increase community resilience. At the response and recovery phases, it can be used as a reactive tool to plan warning, evacuation and recovery needs (Cox et al. 2007; Granger 2003).

Social vulnerability to natural hazards is dynamic (Cutter and Finch 2008). A SoVI can be used to capture the dynamics of the social demographic characteristics at the UBI. This type of work can provide critical insights into the changing nature of the UBI over time. The Australian census was recently undertaken (August 2011) and when the data are released, a comparative analysis could be completed. It would also be interesting to investigate whether there is a relationship between UBI expansion and the level of social vulnerability of the people who move to the UBI. Knowledge about the spatio-temporal dynamics of social vulnerability can also be used to suggest future preparedness and response planning needs.

Nevertheless, social vulnerability is only one component of vulnerability. To have a complete picture, physical factors such as infrastructure and geographic features, community resources, and mitigation and preventive measures that may help to minimize impacts need to be combined. Social vulnerability is a vital component of the risk paradigm as it contributes to determining the level of resilience of particular segments of society and in this instance, particular geographic locations. Therefore, a social vulnerability assessment is an important step in any in future risk and vulnerability studies.

The results of this study emphasize the importance of social vulnerability assessment in an Australian context: different population characteristics are related to vulnerability within populations in different places. Our methodology can be incorporated into existing emergency management plans and used in conjunction with other existing data on vulnerability and risk to understand the vulnerability of a particular community and area. Such assessments can help to develop more effective programs in order to utilize available resources in an efficient way by identifying the most vulnerable populations in a region. Once hazard managers know who is vulnerable, they can determine what specific actions might be best taken to decrease impacts of the hazard on that population.