Evaluating urban fire vulnerability and accessibility to fire stations and hospitals in Austin, Texas

Anthropogenic climate change has increased the frequency and intensity of fires. Despite their widespread consequences, current research has largely overlooked urban fires and their associated vulnerability. This study seeks to identify patterns of fire vulnerability, map out areas with high fire vulnerability and limited access to fire stations and hospitals, and ultimately determine the factors contributing to increased fire incidents. Principal Component Analysis was used to develop a fire vulnerability index comprising variables capturing health status and socio-environmental factors. Enhanced 2-step floating catchment area (E2SFCA) analysis was conducted to determine relative accessibility to resources such as hospitals and fire stations. Ordinary least squares (OLS) regression and geographically weighted regression (GWR) were utilized to determine factors associated with higher fire incident counts. The results of the fire vulnerability analysis highlight areas of high fire vulnerability in the eastern periphery and the north-central parts of Austin. Moreover, the eastern periphery experiences decreased accessibility to fire stations and hospitals. Finally, the results of the GWR analysis highlight a varied negative relationship between health vulnerability and fire incidents and a positive relationship with socio-environmental vulnerability. The GWR model (R 2 : 0.332) was able to predict a greater extent of the variance compared to OLS (R 2 : 0.056). Results of this study underscore that areas with socio-environmental vulnerabilities are likely to face a higher number of fire incidents and have reduced access to hospitals and fire stations. These findings can inform public health officials, city planners, and emergency services departments in developing targeted strategies to mitigate the harm caused by fire incidents.


Introduction
Urban areas around the world are growing at a rapid rate due to increased urbanization.This increased urbanization increases population density within city centers and suburbanization in the peripheral regions.Recently, this increased population density and increased intensity and frequency of fires have highlighted fire as a major hazard.Fire presents a significant risk to human health and safety, primarily through burns and the adverse effects of air pollution caused by smoke.However, the degree of individual vulnerability to fire and smoke-related air pollution can vary and be influenced by various socioeconomic and health factors.In dense urban areas, fires may affect many more individuals due to dense housing.In contrast, fires may burn longer in suburban areas due to closer proximity to the urban-wildland interface [1,2].Fire requires rapid response from fire stations and an understanding of fire vulnerability due to reduced adaptive capacity.
In recent years, much focus has been placed on assessing wildfire risk, especially in rural areas, due to climate change exacerbating existing wildfires.While there has been a significant emphasis on rural areas, it is crucial to highlight the importance of examining urban fire vulnerability and its impacts.Understanding fire vulnerability in urban settings is vital to informing fire stations' planning and response strategies.This understanding is vital in rapidly expanding urban and suburban regions that frequently interface with wildfire-prone rural areas.This holistic approach ensures comprehensive fire preparedness and response strategies for evolving fire threats.
This study seeks to address the following questions: (1) What are the spatial patterns of fire vulnerability within Austin, TX? (2) What is the spatial distribution of areas that have excess fire demand and reduced accessibility to fire stations and hospitals?and (3) What factors are associated with an increased number of fire incidents?

Literature review
The warming climate due to anthropogenic climate change has led to the increased frequency, intensity, and duration of extreme weather events such as extreme heat and drought [3,4].Extreme weather conditions such as drought and heat increase the fire risk as these conditions create more fire-susceptible material [3,5].Several studies have noted this phenomenon, primarily focusing on wildfire risk [6,7].While much of the literature focuses on wildfire risk, many similar conditions, such as hot and dry weather and abundant flammable material, are still found in urban environments.Due to the increased population density of metropolitan areas, fires in these areas may pose more significant health and economic consequences.
It has been widely recognized that fires harm human health.Coming into contact with the fire itself may lead to burns and other trauma, and the smoke caused by fires can worsen air quality and exacerbate breathing problems.In recent years, large wildfires have significantly increased air pollution [8][9][10].With climate change leading to rising temperatures, fires are becoming more frequent and increasingly intense [11].Moreover, the burden of fires on air quality is projected to increase dramatically over the next century due to climate change [10].
This increased burden of fires on human health may lead to various detrimental health effects, such as respiratory symptoms, cardiovascular issues, and increased emergency room utilization [12].These health effects may lead to acute symptoms such as heart failure, stroke, and difficulty breathing that may exacerbate chronic conditions [12].As a result, individuals with chronic health conditions and limited socioeconomic resources are most vulnerable to the consequences of fire.Individuals in these areas need easily accessible fire stations and hospitals, which are essential for preventing fires from escalating and stabilizing patients.
Studies have investigated factors that contribute to vulnerability.However, most research in the field is focused on vulnerability due to wildfires or at the wildland-urban interface (WUI).Giovanni et al. investigated geological factors that may contribute to fire vulnerability.The study's results focused on fuel characteristics and weather conditions as measurable quantities by which an area becomes increasingly at risk for wildfires [13].However, a significant limitation of this approach is that it does not account for ignitions due to human activity.Moreover, weather patterns may be more unpredictable when addressing wildfire risk under climate change conditions.This uncertainty would make a reactive approach to protecting communities from fire ineffective compared to a proactive approach.
In contrast to a focus on wildfires and the environment, an emerging theme in the literature is investigating fire vulnerability on humans at different levels, ranging from global to individual buildings in various locales and socioeconomic contexts.Global approaches to determining fire vulnerability have been employed by Chuvieco et al., who focused on ecological parameters and house values at the wildland-urban interface [14].While this broad approach allows for broader ecosystem conservation approaches, ecological modeling may be difficult for local jurisdictions to undertake, especially in areas that may be low-income or rapidly developing.These constraints at a local level are discussed by Twigg et al., whose work focuses on strategies to determine fire vulnerability in low-income and informal settlements [15].Their work highlights a critical point in the discussion around fire vulnerability and resilience: the need for adequate data on fire incidence and causes and effects.This point highlights the need for fire vulnerability approaches to utilize publicly available data or establish systems to gather data, such as through local governmental institutions or the public.
Moreover, further work investigating fire vulnerability in low-income and informal settlements highlights the need to consider socioeconomic factors that augment fire vulnerability in addition to environmental factors [16].Investigations of fire vulnerability in the wildlandurban interface in southern France focused on vegetation and housing density as risk factors and highlighted the need for accurate fire incident data [17].The authors of this study concluded that isolated housing units in WUIs have the highest vulnerability.These findings may also have implications for more densely populated urban areas undergoing suburban sprawl and highlight the need to consider housing density in characterizing fire vulnerability.Other approaches to characterizing fire vulnerability have focused on high-rise buildings and densely populated urban areas.An example of this is the work of Masoumi et al., who focus on Zanjan, Iran, and utilize unmanned aerial vehicles in combination with data about the building characteristics [18].While this approach provides rich data on fire vulnerability at a building level, this approach may not be feasible for much larger study areas or rapidly developing areas.Similarly, other works focus on buildings and specific building dynamics in older parts of cities in Portugal [19].While this approach does not necessitate using unmanned aerial vehicles, it does require building inspections that may not scale to larger areas or areas that are still developing.Hermawan et al. used a GIS-based approach to characterize fire vulnerability in a densely populated district of Jakarta, Indonesia [20].This approach integrated information about the environmental factors (road class, housing density, housing quality, population density) and social factors (percent of elderly and children, people with disabilities, and population sex ratio) and has been corroborated by other studies investigating fire vulnerability [21].While this approach proves useful in developing a fire vulnerability index, it does not account for health risk factors that may predispose certain individuals to worse outcomes after a fire.
Recent studies have also investigated health risk factors associated with hospitalization and death due to residential fires.Ghassempour et al. investigate different types of fires and health outcomes from residential fires in Australia.Their study utilizes factors such as Age, Sex, Country of Birth, health comorbidities using the Charlson Comorbidity Index, socioeconomic factors through the Socio-Economic Index for Areas (SEIFA), and accessibility through the Accessibility/Remoteness Index for Australia (ARIA+) [22].Findings from this study suggest that comorbidities and sustaining full burns were associated with higher costs and longer lengths of stay [22].Similarly, for mortality, burns and smoke inhalation were critical factors along with ICU admission.This study contains many strengths, namely detailed information on the type of fire and endpoint information regarding health outcomes.Nevertheless, the replicability of this study's design in contexts outside of Australia presents significant challenges.This is primarily because other jurisdictions may lack equivalents to ARIA+ or SEIFA or may not provide these indices at a geographic resolution finer than the national level.Moreover, calculating the Charlson Comorbidity Index requires detailed information about an individual's health, which may not be possible at the population level.As such, a methodology that would be easily adaptable would require publicly available aggregate data.
Rappold et al. incorporate this adaptability by developing a community health vulnerability index using aggregated data.This index used health factors such as county prevalence of various health conditions and socioeconomic status indicators.Using this index, the study's authors determined that more vulnerable counties were at a greater burden of wildfire smoke than less vulnerable counties [23].While a wide variety of approaches are utilized to determine fire vulnerability, few approaches have integrated information about socioeconomic, environmental, and health factors and utilized methodologies that are accessible and scalable for local city planners.
Beyond characterizing fire vulnerability from a vulnerability perspective, another major challenge in facilitating resilience is improving accessibility to hospitals and fire stations, which act as resources to help cope with fires.Accessibility is another central theme within the literature.Shahparvari et al. use a location-allocation modeling approach to determine the ideal location for new fire stations.This approach utilizes information about road density, traffic light density, fire station distance, and fire incidents to determine new locations for fire stations to improve response time [24].Other studies, such as by KC et al. and Mao et al., utilize two-step floating catchment area methods (2SFCA) [25,26].Both studies utilize historical fire incident data to determine accessibility to fire stations and highlight the 2SFCA methodology as a means of determining spatial accessibility to resources.Other studies in the literature build upon this methodology and use the enhanced 2SFCA (E2SFCA) approach, which accounts for distance decay and the behavior of individuals to potentially travel further for resources than what may be available in a particular catchment area [27][28][29].This methodology has been utilized in a variety of different healthcare settings, such as measuring accessibility to COVID-19 testing resources [28], HIV testing and treatment services [30], and hospitals [29,31].These studies highlight the versatility of the E2SFCA methodology for determining accessibility to a variety of healthcare and fire relief resources.However, a significant limitation of these studies is that they focus on optimizing resource allocation without consideration of variance in fire vulnerability, which may lead to different demand characteristics in the future.
This paper develops a novel fire vulnerability index using publicly available data to capture information about health and socio-environmental factors at the census tract level.Additionally, this paper investigates accessibility to resources such as fire stations and hospitals that may relieve jurisdictions experiencing fires.Then, we suggest public policy and public health strategies to develop a resilient system to respond to fire incidents within urban environments.

Study location
The study location was restricted to the city of Austin within Travis County, Texas.Due to multiple fire services providers operating within the city of Austin, only census tracts within Travis County and part of Austin Fire Department's jurisdiction were included (n = 140 of 218).The final study area was created using a spatial join and is depicted in Fig 1 .Austin Fire Department (AFD) is the largest fire services department in Austin and maintains extensive records of fire incidents.Moreover, the city of Austin is experiencing tremendous growth, especially in the eastern and southern parts of the city.This growth has led to concerns about urban sprawl and insufficient infrastructure to accommodate the increasing demand for public services and the capacity of fire services to protect rapidly growing populations [32].

Data sources
Fire incident data.Fire incident data was collected from the City of Austin online data portal.Incidents were collected from 2016-2020 (n = 21,344) and spatially joined into census tract boundaries to provide a cumulative number of fire incidents per census tract.The data on the number of fire incidents underwent a natural logarithm (ln) transformation to provide a normalized distribution.Socio-environmental data.Socio-environmental data was collected from the 2016-2020 American Community Survey (ACS) 5-year estimates.In the initial phase of our study, we selected variables based on their relevance to social vulnerability in the context of environmental hazards and disaster management, ranging from general disasters to specific risks, including wildfires/urban fires.We began by reviewing indicators utilized in Cutter's [33] and the CDC's methodologies, outlined by Flanagan et al. [34], supplemented by contributions from other researchers such as Bixler et al. [35], Blaikie et al. [36], Inostroza et al. [37], Rivière et al. [38], Schmidtlein et al. [39], and Tate [40].The variables we initially considered include: • Race and ethnicity (Blaikie et al. [36], Cutter et al. [33], Flanagan et al. [34], Peacock, Morrow, and Gladwin [41,42], Tate [40], Schmidtlein et al. [39]).
From an initial set of seventeen socio-environmental variables, four variables addressing limited English proficiency, poverty levels, disabilities, and unemployment were excluded due to data availability.Through applying Principal Component Analysis (PCA) and considering the CDC's health indicators, we further refined the list by excluding seven additional variables -Hispanic, receipt of social benefits, living alone, lack of internet access, median household income, housing size, and population density.In relation to health indicators, the PCA analysis left us with six variables (Black, elderly, children, housing unit, old home ratio, no HVAC facilities) to construct the fire vulnerability index at the census tract level.
Health data.Health condition data was collected from the Population Level Analysis and Community Estimates (PLACES) dataset from the Centers for Disease Control and Prevention (CDC).For this analysis, the CDC 2020 dataset was utilized.Due to the 2020 CDC PLACES dataset utilizing the 2010 census tract boundaries, all analyses were done utilizing those census tract boundaries.Our analysis focused on broader area trends and general public health implications rather than specific census tract-level conclusions in an attempt to address this limitation.This approach ensures that our findings remain relevant and applicable despite potential shifts in census tract boundaries.

Fire vulnerability analysis
Principal component analysis.We have adopted the principal component analysis (PCA) technique from Cutter's Social Vulnerability Index (SVI) to consolidate various variables into a single fire-vulnerability index by determining principal components (PC) of synergistic variables.Cutter's SVI is notably theoretical and flexible and is often tailored to specific research questions and applications [33].This flexibility allows for the inclusion of a diverse array of variables while making it versatile for academic studies [33].The PCA analysis helps identify clusters of covariant factors represented by a principal component, thus reducing several variables into a singular index [45,46].
The following health variables were selected for inclusion in the original PCA analysis: pressure medications, the prevalence of cancer, asthma, high blood pressure, coronary heart disease, smoking, diabetes, obesity, high cholesterol, and stroke frequency of regular checkups and taking blood pressure medications, and access to healthcare.
The following socioeconomic variables were originally considered for inclusion in the PCA analysis in addition to the variables used in the final analysis (Table 1): percent non-Hispanic White, Hispanic, living alone, with no access to a computer or internet, low economic status (receipt of social benefits), limited English proficiency, poverty levels, disabilities, unemployment, median household income, housing size, total population, percentage of houses built before 1980, percentage of adults with high cholesterol, no leisure-time physical activity and current lack of health insurance.From both socio-environmental and health data, fifteen variables were omitted from the analysis, leaving a total of 18 variables used to create a health index and a social vulnerability index, which then were used to create a fire vulnerability index.Components were selected based on having an eigenvalue greater than one and a factor loading value exceeding 0.30, considering our sample size of 140.One exception was made for the crude prevalence of asthma, considering that smoke effects may exacerbate asthma conditions [47].Table 1 identifies how these variables were sub-categorized.Only the prevalence of checkups was orientated negatively, while all other variables were orientated so that increased values indicated increased fire vulnerability.This one variable was adjusted by negation to ensure the same cardinality for all variables in the index.
Utilizing the CDC social vulnerability index methodology, all variables were assigned a percentile rank (ranging from 0 to 1) [34,48].Variables in the bottom 10% (� 0.1) or top 90% (� 0.9) were floored to 0 and 1, respectively.The rank values of all variables within each PC were added together.Then, each PC was transformed to yield a Z-score and added to each other to create the Health Index (PCs 1 + 2) and Socio-environmental Index (PCs 3 + 4).These indices were transformed to yield a Z-score, added together, and normalized to create the final fire vulnerability index.In sum, our methodology harmonizes core principles from both the CDC and Cutter's SVI while introducing specific modifications and enhancements to address the unique aspects of fire vulnerability.
Correlation analysis.After developing the indices, a correlation analysis was conducted to determine if there were any significant associations between highly vulnerable areas and fire incident counts.The extent of correlation was determined using Pearson's correlation coefficients (r) and corresponding p-value.
Spatial autocorrelation analysis.Spatial autocorrelation analysis was conducted to determine if there was a significant grouping of fire incident counts, health vulnerability, or socioenvironmental vulnerability.For this aim, Moran's I analysis enables the prediction of spatial variation and the extent of clustering.Local indicators of spatial autocorrelation (LISA) illustrate the extent to which significant clustering surrounding a particular census tract occurs as a proportion of the global indicator of spatial autocorrelation (Moran's I) [49].The similarities and differences between a census tract and neighboring tracts were determined using LISA for each of the three variables.High-high (HH) and low-low (LL) indicate concordant clustering of high-or low values, respectively.High-low (HL) or low-high (HL) indicate clustering of discordant values.
Regression analysis.A linear regression model was used to identify a relationship between the dependent variable (fire incidents) and independent variables (health and socioenvironmental vulnerability) throughout the study area.Initially, the variables composing the indices were utilized for the regression analysis.However, most variables failed the collinearity test and mostly had a VIF greater than 10.As a result, the indices were used for the regression analyses.The dataset meets the assumptions of the ordinary least squares regression model as the residuals are independent and normally distributed with a mean of zero and constant variance.
A geographically weighted regression (GWR) model expands upon the linear regression model by allowing model parameters to vary across space.The GWR model attempts to determine the relationship between the independent and dependent variables at each census tract included in the study.This model assumes that parameters colocated near each other have a stronger influence on each other relative to parameters for census tracts located far from each other.The extent of influence or weight is determined using a distance decay function centered around a particular observation, i.
Sensitivity analysis was conducted to select the optimal configuration.The results of this analysis are in S1 Table .Optimizing for the lowest AICc, highest R 2, and Adjusted R 2 , a model with an adaptive kernel and AICc search criteria was selected.The coefficients of each variable were mapped using the Natural Breaks method, and an additional break was added to distinguish between sign changes.
Fire service accessibility analysis.The enhanced two-step floating catchment area (E2SFCA) method was utilized to determine the accessibility of each census tract to nearby hospitals and fire stations.The E2SFCA method has commonly been used to characterize accessibility to various resources, including hospitals [25,26,50,51].In this study, two instances of this analysis were conducted: one evaluated accessibility to hospitals using population as the demand metric, while another evaluated accessibility to fire stations.We modeled the demand side of the calculation based on aggregating data from the entire census tract due to the excessive computational demand of calculating three isochrones and catchment area matrices for 13,943 fire incidents.As such, the centroid of each census tract was utilized as the demand location.On the supply side, hospital location data was obtained from OpenStreet-Map [52], and fire station data was obtained from the City of Austin [53].Upon subsequent review of the hospital data, a few private clinics were found to be misclassified as hospitals.However, these clinics were located far outside the study area and thus did not affect the analyses.No data regarding their relative capacity has been published for fire stations and hospitals.As a result, all supply-side locations have been modeled as having the same supply.
As the principal mode of transportation in Austin, TX, is by automobile, and EMS and fire trucks can be modeled as automobiles, travel distances have been modeled as automobiles.Certain thresholds were determined using the HERE isoline API to create catchment areas for 10 minutes and 20 minutes.These thresholds were selected as national standards for EMS response time, which is around 5 minutes for 90% of the time, and these thresholds capture this target as well as 2-4 times the response time [54].
Using the catchment areas, a catchment matrix was created that indicates what travel time each hospital has from each census tract.As a result of this calculation, a matrix is developed with each census tract and the travel time to each hospital within categories of 0-10 minutes, 10-20 minutes, and over 20 minutes.
A weighted matrix takes the information from the catchment matrix and assigns greater weight to closer combinations.This was done using a Gaussian impedance function, and the following weights were assigned: 1 for 0-10 minutes, 0.13 for 10-20 minutes, and 0 for over 20 minutes.
The demand vector is calculated using the total population of each census tract or the number of fire incidents.Multiplying the demand vector by the weight vector produces a vector (weighted demand) that can be modified to account for the different supplies and resources of each fire station or hospital.However, this modification is equal to 1 for all locations due to a lack of data on the capacity of each hospital or fire station.
Multiplying the weighted demand vector by the weights again produces the Spatial Access Index (SPAI).This index is highly specific to a particular analysis and cannot be interpreted without contextualizing these values within the broader region.Normalizing SPAI by dividing all of them by the average SPAI in the area produces the Spatial Access Ratio (SPAR).This value ranges from 0 to 2, with 1 indicating that a particular census tract has exactly average access, while values above 1 indicate greater accessibility.
Finally, bivariate mapping of the study area was utilized to determine which census tracts experienced high fire incidents and low accessibility to hospitals and fire stations.Bivariate maps were developed using the ArcGIS Pro.Bivariate mapping utilizes the quantile method for categorizing data into various baskets that each display a particular color.greater health vulnerability.There were a few interspersed pockets of high total fire vulnerability, mainly located around the eastern periphery and a few pockets in the north-central and southeastern parts of Austin.The results of the fire vulnerability mapping indicate that the rapidly growing eastern part of Austin is increasingly vulnerable to fires due to health and socioeconomic vulnerability.On the other hand, the western periphery is more vulnerable from a health aspect but may have more resources to cope with the challenges of fire.

Fire vulnerability assessment
Correlation analysis.While the correlation between fire incident count and socio-environmental vulnerability is statistically significant, the correlation is relatively weak (r = -0.174,p < 0.05) (Table 2).This contrasts with other findings that suggest that factors that compose socio-environmental vulnerability (such as race, unemployment, and elderly status) may be associated with higher numbers of fire incidents [55].Due to the contradictory evidence, further evidence is needed regarding whether socioeconomically disadvantaged communities are more prone to fire incidents.vulnerability types overlap with high fire incident counts in portions of the eastern and northsouth periphery.Census tracts with high fire incident count and low vulnerability tend to be located near census tracts with the same high fire incident count exposure but with high vulnerability.Census tracts with high socioeconomic vulnerability and fire incident count are dispersed throughout the study area.Notably, there is high health vulnerability in the western periphery, yet there is a relatively low number of fire incidents.Spatial autocorrelation analysis.Results of the Global Moran's I analysis (Fig 5) indicate a statistically significant spatial autocorrelation in the study area, suggesting the tendency for similar or dissimilar values to be clustered, dispersed, or randomly distributed across the geographic space.This finding is significant for the normalized fire incident count (Global Moran I index (0.31), p < .00)and health vulnerability (Global Moran I index (0.39), p < .00)but not significant for the socio-environmental vulnerability index, meaning that the number of fire incident and health vulnerability have a positive tendency for spatial clustering at the census tract level.Meanwhile, the spatial pattern of the socio-environmental vulnerability index could be random.

Bivariate mapping. Fig 4 illustrates the relationship between health vulnerability and socio-environmental vulnerability and the number of fire incidents. Generally, both
Local indicators of spatial autocorrelation (LISA) provide insight into spatial clusters on a local scale (Fig 6).Census tracts with a high level of a variable clustered around other similar census tracts are indicated as HH (High-High) and similarly for other permutations (LH, HL, LL).For fire incidents, there is a notable HH cluster in the south-central and north-central parts of Austin, while an LL cluster extends from the west into the northern part of the study area.Clusters of high-high health vulnerability are scattered throughout the study area.Still, they are mainly centered around East-Central Austin, while the main low-low vulnerability cluster is in West-Central Austin.A similar pattern of high-high clustering exists for socioenvironmental vulnerability to a lesser extent.
Regression analysis.A regression analysis was conducted on the influence of health and socioeconomic vulnerability index along with accessibility on the number of fire incidents.Ordinary least squares regression (OLS) was used to determine global influence, while Geographically Weighted Regression (GWR) was utilized to determine if the extent of influence varied throughout the study area.Results from both analyses are located in Table 3 and Fig    Generally, health vulnerability is likely to impact fire incident counts negatively, while socio-environmental vulnerability positively influences fire incident counts.However, some census tracts are exceptions to this observation, highlighted in red and purple in Fig 7 .For health vulnerability, census tracts in north-central Austin are likely to influence fire incident counts positively, while for socio-environmental vulnerability, a similar inverted coefficient is found in the northeastern portion of Austin.Generally, the maps indicate that the southern and eastern periphery of Austin experience low accessibility to hospitals and fire stations while also experiencing a relatively high number of fire incidents.On the other hand, the western periphery generally has poorer access to hospitals and fire stations, yet there is a relatively lower number of fire incidents.The central-north region of Austin experiences a relatively high number of fire incidents but has relatively high accessibility to hospitals and fire stations.A few census tracts in the northeast have relatively high fire incidents and low accessibility to fire stations, yet they have greater accessibility in terms of hospitals.

Demand-side analysis: fire vulnerability assessment
The map of the fire vulnerability index within Austin indicates that different parts of the census tract have different needs in terms of fire vulnerability.There are a few census tracts that are vulnerable from both a socioeconomic and a health perspective.However, most census tracts tend to be vulnerable in one index or another.The central area may not have health conditions that exacerbate fire vulnerability but also may not have adequate resources to adapt to fires.For this area, mitigation efforts could be placed on increasing community resilience to fires through economic grants and other methods to improve environmental factors to make neighborhoods less vulnerable to fire incidents.Potential strategies could include implementing a modified Haddon Matrix, as proposed by Twigg et al. [15], to engage the community in understanding the systemic causes of fires in neighborhoods that have high socio- environmental vulnerability.For the periphery, a different approach focused on health improvement may be more vital as these census tracts experience greater vulnerability from a health standpoint.Policymakers, community partners, and fire departments may implement varied approaches to reducing fire vulnerability in different parts of the Austin area.
Only the socio-environmental vulnerability component of the fire vulnerability index is slightly positively correlated with the total number of fire incidents (r = 0.173, p < 0.05).Socio-environmental vulnerability is also negatively associated with the distance to the nearest hospitals, indicating that those census tracts may have to travel further for hospital care.(r = -0.183,p < 0.05).Overall, the correlation data is relatively weak, and further investigation may need to be conducted on the relationship between socio-environmentally disadvantaged communities and the number of reported fire incidents.If such a relationship exists, targeted educational initiatives led by local fire departments and community leaders could significantly mitigate the frequency of fire incidents in these census tracts.Additionally, implementing strategies such as advanced sensor technologies and alert systems could serve as effective early warning mechanisms, enabling quicker mobilization of resources and bolstering overall community resilience [56].
The findings from the demand bivariate mapping highlight a significant need in the central-eastern region, particularly from a health vulnerability perspective.Addressing this issue may include the implementation of public health campaigns and other strategies to reduce the incidence of chronic diseases in this area.The demand from a socio-environmental vulnerability perspective is less clustered beyond a significant cluster in the eastern and north-central portions.The lack of clustering may pose a challenge for interventions as such interventions may not be spatially concentrated.This dispersion indicates the need for efficient allocation of resources as areas of high demand are clustered together but may have differences in socioenvironmental vulnerability.The overlapping regions in the east and north-central portions, which have high vulnerability in both aspects and high fire incident counts, suggest a dire need for a multidisciplinary approach to reduce fire incidents and improve community resilience against fires.
Another critical area for interventions to target would be areas with high vulnerability but a relatively low number of fire incidents.In the future, such areas may experience an increased number of fire incidents due to climate change.For these areas, such as West Austin, preventative methods such as improving community resilience and education campaigns can help mitigate future fire vulnerability.
Results from the spatial autocorrelation analysis highlight distinct clusters of high fire incident counts in the north and south-central portions of Austin.In contrast, significant cluster exits of low fire incident counts exist in the northwestern periphery.Further investigation may be required to determine what factors contribute to the increased number of fire incidents in the central portions of Austin and what contributes to the decreased number of fire incidents in the western portion.High health vulnerability is primarily clustered in the eastern portion of Austin, while there is a significant low vulnerability cluster in the western portion.Part of this could be due to demographic differences between east and west Austin and relative accessibility to hospitals and other medical facilities.Notably, for socio-environmental vulnerability, clustering patterns tend to be similar to health vulnerability, except with a greater frequency of highly vulnerable census tracts being located adjacent to much less vulnerable census tracts.This finding could suggest the need to investigate characteristics of neighborhoods on a scale much smaller than census tract, such as at the block group level.Notably, a few clusters of high-low socioeconomic vulnerability exist and may highlight the need for highly targeted interventions.In broad terms, the results of the spatial autocorrelation analysis indicate significant clustering of vulnerability and fire incident counts throughout the study area and inform investigating trends on a smaller scale.
Results from the GWR highlight that health vulnerability may be associated with fewer fire incidents, while socio-environmental vulnerability may be linked to increased fire incidents.The exact mechanisms behind this relationship require further investigation.Counterintuitively, the coefficient is inverted in the north-central Austin area.Reducing health vulnerability may also provide additional benefits in lowering fire incident counts in this area.Generally, the health vulnerability analysis results highlight that while increases in health-vulnerable populations may reduce fire incidents, those populations are still highly vulnerable to the consequences of any fires that do occur.Furthermore, the strength of this association varies throughout the study area, generally weakening in the eastern periphery and a census tract in the southwest periphery of Austin.Reasons for this variation are currently unknown but could be due to these areas having different levels of urbanization or other environmental factors such as industrial land use, air quality, and access to green spaces.
Socio-environmental vulnerability generally has a greater influence in the western portion of Austin, an area that has historically been more socioeconomically advantaged.Explanations of this observation are supported by other literature suggesting socio-environmental vulnerability as a predictor of fire incidents [55].
A portion of northeastern Austin has a negative coefficient for socio-environmental vulnerability, implying a protective factor against fire incidents.This finding is counterintuitive and may require further investigation.While a portion of this area overlaps with the inverted area in health vulnerability, this area with a negative socio-environmental vulnerability GWR coefficient extends further to the north and east.Overall, these findings indicate that there may be merit in focusing on vulnerable households surrounded by less vulnerable households to reduce fire incident counts and broadly improve socio-environmental conditions.In addressing health vulnerability, strategies should emphasize cultivating individual and community resilience to fire incidents.This can be achieved through health promotion campaigns, which may include advocating for using air filters and other relevant measures.

Supply-side analysis: Fire service accessibility
The results of the bivariate mapping indicate that the southeastern and northeastern periphery should be prioritized in building more fire stations as these census tracts have low accessibility to fire stations and a relatively high number of fire incidents.Especially as these areas are rapidly growing due to their relatively lower cost of living compared to the rest of Austin, this unmet need may only increase in the future.While portions of the western periphery experience low accessibility as well, they also experience a relatively low number of fire incidents.However, these census tracts should also be prioritized as they may be highly vulnerable if a fire incident occurs due to their high health vulnerability and low accessibility to fire stations.In terms of hospitals, the southern portion of Austin experiences generally lower accessibility while the southeastern portion also experiences a relatively high number of fire incidents.Mitigation strategies can include building medical facilities in the southeastern portion that are well-equipped to handle fire risks.
For both fire stations and hospitals, the north-central regions of Austin experience high accessibility and many fire incidents.Thus, strategies to safeguard these communities should prioritize fire prevention and effective management of existing facilities, given that the area appears to have adequate access to local hospitals and fire station resources.

Limitations
Several limitations exist within this study, ranging from limitations due to the data availability to data analysis.Regarding data availability, this study is limited by investigating only the area that intersects Travis County, TX, and the Austin Fire Department's service area.Thus, there may be areas that other agencies serve that the analysis may not include.It is unknown whether both AFD and outside agencies serve these areas, so it may not be clear what the impact of this limitation may be on the findings.
Another limitation is that the fire incidents were analyzed in aggregate over five years.As a result, shifting patterns in the distribution of fire incidents over time may be masked.This may be especially important as Austin has experienced rapid population growth in the past five years.Future work may investigate shifting patterns of fire incidents as an urban area experiences rapid growth and development.Likewise, the analysis also considers every fire incident to be equivalent as there is no data collected thus far on the intensity of the fire.Similarly, housing factors such as the material and type of housing (single, multiple-family, or apartments) may modify the vulnerability of residents to fire.Future analyses may investigate vulnerability by considering more factors than just the age of housing.
Another broad set of limitations stems from accessibility analysis.Due to computational resource constraints, all fire incidents were aggregated into the center of their corresponding census tract rather than being considered by their original location.Although the difference should be canceled out within a census tract by using the centroid, this may misrepresent the true accessibility, depending on whether there is a higher population density in one part of the census tract versus the other.Moreover, the accessibility analysis considers all hospitals and fire stations to have functionally the same capacity due to a lack of data on individual hospital or fire station capacity to address fires.As a result, two fire stations with vastly different resources may be considered equivalent and, likewise, with hospitals that may differ in their capacity to treat acute fire-related medical issues.Furthermore, psychiatric hospitals or other specialty care hospitals were also considered to be hospitals on the basis that they would have the capacity to provide emergency life support in the case of an emergency but may result in misleading information about accessibility if they are clustered around general hospitals.
Finally, our analysis produced weak correlational evidence, which may need further investigation to validate and determine if there is a significant relationship between socioeconomic vulnerability and fire incidents.

Conclusion
This paper introduces a novel fire vulnerability index focused on urban fires instead of wildfires.Moreover, an accessibility analysis was conducted to determine which areas experience increased fire incidents and decreased accessibility.Findings from our correlation analysis suggest that socio-environmentally vulnerable regions may experience more fire incidents and lower accessibility to hospitals.Our accessibility analysis highlights the need for improved accessibility to hospitals and fire stations in the rapidly growing eastern and southern periphery.Finally, our regression analysis highlights the differing influence of vulnerability on fire incidents throughout the city, including areas with inverted relationships between vulnerability and fire incidents.While further investigation is needed, these findings can inform city planners, fire departments, and emergency medical services to improve Austin's resiliency toward fire incidents.
Fig 2 below indicates the distribution of fire incidents across census tracts.

Fig 8
Fig 8  illustrates the results of the accessibility analysis and maps the distribution of accessibility to hospitals or fire stations based on population versus fire count (natural log-transformed).Generally, the maps indicate that the southern and eastern periphery of Austin experience low accessibility to hospitals and fire stations while also experiencing a relatively high number of fire incidents.On the other hand, the western periphery generally has poorer access to hospitals and fire stations, yet there is a relatively lower number of fire incidents.The central-north region of Austin experiences a relatively high number of fire incidents but has relatively high accessibility to hospitals and fire stations.A few census tracts in the northeast have relatively high fire incidents and low accessibility to fire stations, yet they have greater accessibility in terms of hospitals.