How Vulnerable Are American States to Wildﬁres? A Livelihood Vulnerability Assessment

: Quantifying livelihood vulnerability to wildland ﬁres in the United States is challenging because of the need to systematically integrate multidimensional variables into its analysis. We aim to measure wildﬁre threats amongst humans and their physical and social environment by developing a framework to calculate the livelihood vulnerability index (LVI) for the top 14 American states most recently exposed to wildﬁres. The LVI is computed by assessing each state’s contributing factors (exposure, sensitivity, and adaptive capacity) to wildﬁre events. These contributing factors are determined through a set of indicator variables that are categorized into corresponding groups to produce an LVI framework. The framework is validated by performing a principal component analysis (PCA), ensuring that each selected indicator variable corresponds to the correct contributing factor. Our results indicate that Arizona and New Mexico experience the greatest livelihood vulnerability. In contrast, California, Florida, and Texas experience the least livelihood vulnerability. While California has one of the highest exposures and sensitivity to wildﬁres, results indicate that it has a relatively high adaptive capacity, in comparison to the other states, suggesting it has measures in place to withstand these vulnerabilities. These results are critical to wildﬁre managers, government, policymakers, and research scientists for identifying and providing better resiliency and adaptation measures to support states that are most vulnerable to wildﬁres.


Introduction
Wildfires are crucial for ecosystem dynamics by balancing fuel types and creating appropriate vegetation for maintaining healthy forested regimes [1]. Despite the integral ecological role of wildfires, uncontrolled burns can cause widespread environmental, economic, social and sustainable development impacts [2][3][4]. Such wildfire impacts include losses to human lives; incurring financial losses from buildings and homes; widespread social, health and economic costs through evacuations, smoke exposure, and loss of tourism revenue [5][6][7]. The Insurance Information Institute, gives an example of financial loss due to wildfires, including the 2019 wildfires in California and Alaska that created a loss of 4.5 billion dollars in damages, largely resulting from the California Kincade and Saddle Ridge wildfires. In order to minimize ignition and spread during this time, California's electrical utility provider issued rolling blackouts to homes and businesses during high wind and extreme dry conditions. However, this inevitably cost the state billions of dollars in losses [8]. It is therefore evident that wildfires have a direct impact on the livelihood of many residents in fire-prone communities within the United States, making them vulnerable to wildland fire exposure [9].
Besides economic impacts, changes in social and climate conditions can significantly affect fire regimes, producing greater potential damage than those previously thought [2]. Social factors, such as the expansion of the wildland-urban interface (WUI) (where human settlements, buildings, and wildland vegetation meet), have influenced the dramatic Panel on Climate Change (IPCC) (IPCC, 2001). Exposure represents the magnitude and duration of the climate-related exposure (in our case wildfires), while sensitivity describes the degree to which a system is affected by the exposure, and adaptive capacity describes the system's ability to withstand or recover from the exposure [17,25].
The LVI uses multiple indicators that are aggregated into the IPCC's three contributing factors to produce a vulnerability framework. Studies have applied the LVI method, such as [26] to assess farmers' livelihood vulnerability to global changes in irrigation agricultural practices in Spain. They show that an increase in the adoption of irrigation practices have increased the short-term adaptive capacity while displacing small-scale farming. Studies, such as [27], have also used the LVI approach to assess the livelihood vulnerability of flood risks to farmers for different regions in Indonesia. Results indicate that regions with similar physical characteristics and agricultural dependencies show similar vulnerability levels.
A study [28] used LVI to quantify the vulnerability of communities to heavy lake effect snowfall in the Northwest Territories of Canada. They found that extreme precipitation makes some lake-rich communities more vulnerable than communities farther inland. Therefore, it is acknowledged that there are numerous interpretations on how best to apply exposure, sensitivity, and adaptive capacity concepts to quantify vulnerability [17,25,[29][30][31][32], with key differences among studies that include methods used for scaling, gathering, grouping, and aggregating indicator variables [18].
We adopt an LVI approach, similar to the original methods proposed by [18], to evaluate recent wildfire impacts in the US. This is conducted by developing a framework that combines a set of indicator variables into their respective contributing factors to determine the critical biophysical and human dimension components influencing the livelihood vulnerability of selected wildfire prone states. The information gained from this assessment will provide a clearer understanding as to which states are most vulnerable to wildfires despite their level of wildland fire exposure. This information will be critical to researchers, government organizations, and policymakers for identifying, allotting, and providing better resiliency and adaptation measures, such as aiding in financial, environmental, and social support for states that are most vulnerable to wildfires.

Data and Methodology
Assessing the LVI to wildfires across selected American states are conducted in two folds. First, we develop a framework comprising a set of biophysical, social, and economic factors that is used to assess each state's livelihood vulnerability to wildfires. We acknowledge that our framework provides one possible way of developing a livelihood vulnerability model, and that results could differ depending on the subjective allocation of each indicator variable in our framework. For these reasons, we also conduct a principal component (PCA) analysis to determine the validity of our framework. Second, we calculate the LVI and its contributing factors for each state. We further conduct a sensitivity test to provide additional certainty that our framework is valid, and our results are robust.

Building the LVI Framework
The definitions of the livelihood vulnerability terms used in our framework are summarized in Table A1, which describes the overarching contributing factors comprising exposure, sensitivity, and adaptive capacity (color coded red, blue, and green, respectively). While we acknowledge that the definition of the vulnerability terminologies can differ across disciplines, we develop our framework and conduct our livelihood vulnerability assessment based on the definitions in Table A1, to ensure terminology and interpretation consistency throughout this study. These contributing factors are divided into major components (first level of divisions within each contributing factor). These major components are further divided into sub-components (second level of divisions within each major component) and subsequent indicator variables (measurable units of data for each sub-component) ( Figure 1). are further divided into sub-components (second level of divisions within each major component) and subsequent indicator variables (measurable units of data for each subcomponent) (Figure 1).

Figure 1.
Description of the framework developed for the LVI (box 1 and the central gray circle). LVI is represented by contributing factor (box 2). The contributing factors are sensitivity (blue), exposure (red), and adaptive capacity (green). The contributing factors are further divided into major components (box 3). The major components are color-coordinated with the contributing factors. The major components for sensitivity (blue) are demographic, ignition causes, and environmental index (light blue); for exposure (red) are wildfire occurrence, topography, weather, weather extreme events (light red); for adaptive capacity (green) are social network, natural, physical, human, and financial capital (light green). Major components are divided into sub-components (box 4) and represented by the sub-components in the outermost part of the circle. The sub-components are further divided into indicators (box 5) and not shown in this figure. Refer to Table A2 for each indicator variable.
In our study, the exposure factor pertains to wildfire danger and the physical processes that influence the intensity and severity of fire behavior. Major components within exposure are wildfire occurrence, topography, weather, and extreme weather events. In our framework, an indicator variable under exposure is interpreted as variables that can adversely affect the exposure of wildfire risk to people within a state. For example, "total acres burnt due to wildfires" is an indicator variable that represents how burnt soil disturbs hydrologic and soil conditions leading to increased likelihood of flooding, runoff, and debris flow [33]. This variable can be interpreted as, the greater the number of acres burnt, the greater the exposure humans have to "knock-on" natural hazards, such as flash flooding and landslides within a state (Table A2). Description of the framework developed for the LVI (box 1 and the central gray circle). LVI is represented by contributing factor (box 2). The contributing factors are sensitivity (blue), exposure (red), and adaptive capacity (green). The contributing factors are further divided into major components (box 3). The major components are color-coordinated with the contributing factors. The major components for sensitivity (blue) are demographic, ignition causes, and environmental index (light blue); for exposure (red) are wildfire occurrence, topography, weather, weather extreme events (light red); for adaptive capacity (green) are social network, natural, physical, human, and financial capital (light green). Major components are divided into sub-components (box 4) and represented by the sub-components in the outermost part of the circle. The sub-components are further divided into indicators (box 5) and not shown in this figure. Refer to Table A2 for each indicator variable.
In our study, the exposure factor pertains to wildfire danger and the physical processes that influence the intensity and severity of fire behavior. Major components within exposure are wildfire occurrence, topography, weather, and extreme weather events. In our framework, an indicator variable under exposure is interpreted as variables that can adversely affect the exposure of wildfire risk to people within a state. For example, "total acres burnt due to wildfires" is an indicator variable that represents how burnt soil disturbs hydrologic and soil conditions leading to increased likelihood of flooding, runoff, and debris flow [33]. This variable can be interpreted as, the greater the number of acres burnt, the greater the exposure humans have to "knock-on" natural hazards, such as flash flooding and landslides within a state (Table A2).
Sensitivity describes the degree to which each state is affected by wildfires. Its major components are demographic, ignition causes, and selected environmental indices. The indicator variables under sensitivity are interpreted as variables that contribute to a state's sensitivity to wildfires. For example, the "number of houses within a wildland urban inter- face zone" is an indicator variable under sensitivity because WUI are high-risk wildfire regions due to their accumulation of wildland vegetation, concentration of flammable human structures, and potential ignition sources from sparks left by human activities [34,35]. Therefore, this indicator is interpreted as follows: States with larger WUI area and greater number of homes within the WUI will be at increased risk and sensitivity to wildfires (Table A2).
Adaptive capacity describes the ability of each state to withstand or recover from wildfires. The major components of adaptive capacity include natural capital, physical capital, human capital, social network, and financial capital. The indicator variables under adaptive capacity are interpreted as variables that can positively contribute to the state's mitigation and adaptive strategies for wildfires. For example, "median household income" is an indicator variable under adaptive capacity that can be interpreted as: States with higher income may have more financial resources and financial capital to invest in home and community hardening, thereby having higher mitigation and adaptation capacities (Table A2).
When interpreting the indicator variables under each contributing factor (exposure, sensitivity, and adaptive capacity), usually states with higher values would represent greater exposure, sensitivity, or adaptive capacity to wildfires. For example, states with higher temperatures would have a "greater exposure to wildfires". However, for certain indicator variables, the opposite is true, such as precipitation. States with higher amounts of precipitation will have a "lower" exposure to wildfires. Indicator variables that are interpreted as such require the inverse of their value to be integrated into the LVI equation (e.g., instead of 12 mm of rain, it would be ( 1 12+1 ). These inverse indicators are denoted with an asterisk (*) in Table A2. Please refer to Table A2 for a comprehensive overview of our LVI framework that outlines all the major components, sub-components, and indicator variables used in each contributing factor, along with detailed rationales and interpretation on how we apply each indicator variable to assess the livelihood vulnerability of each state.

Input Variables
The LVI analysis is conducted solely for 14 fire prone American states that are most at risk to wildfires. The states selected are Arizona, California, Florida, Idaho, Montana, Nevada, New Mexico, Oklahoma, Oregon, Utah, Washington, and Wyoming because they experienced the highest risk of wildfires in 2018, as determined from by the maximum acres burnt in 2018 and 2019 and as documented in the NIFC 2019 Wildfire Risk Report (Table A3 in the Appendix A). The 14 states analyzed in this study had the largest acreage burnt in 2018 across the US ( Figure 2). For these reasons, we limit our analysis to comparing the LVI for only the top states most exposed to wildfires. The remainder of the states are not as exposed and will inevitably provide irrelevant comparisons. Though Alaska was included as a top state listed in the 2019 Wildfire Risk Report, it was excluded from our study due to the lack of spatial and temporal comprehensive data, such as those required under sensitivity (e.g., number of houses within the WUI zone), and if included, would have impeded our comparison analysis among the other states.
Our analysis is conducted to determine the current LVI and not future LVI projections. Therefore, most of the data gathered for our assessment were acquired within the past decade (2010-2019). The exception is given to certain indicator variables that represent a long-term climatological average (1950 to 2019). In addition, the elevation data for each state was acquired from 1980, with the understanding that the elevation of each state is not time sensitive and would not have changed drastically if the measurements were acquired in 2019. The year in which the data was acquired for each indicator variable in our framework is indicated in Table A2.
Furthermore, most of the data acquired are entered directly into the framework as raw values, meaning that they did not require additional computations before the LVI was calculated. However, some indicator variables under exposure, sensitivity, and adaptive capacity required further processing to be amenable and included in the analysis. Indicator variables under the exposure that required initial computations included annual average wind speed, humidity, annual precipitation, number of days with higher than 0.1 inches or more of precipitation, and annual temperature. The National Center for Environmental Information (NCEI) provides annual averages of each indicator for various weather observation stations located in each state. The values for every available weather observation station within each state were spatially averaged over the state and temporally averaged over the 1950 to 2018 period before being used in our LVI calculations. Our analysis is conducted to determine the current LVI and not future LVI projections. Therefore, most of the data gathered for our assessment were acquired within the past decade (2010-2019). The exception is given to certain indicator variables that represent a long-term climatological average (1950 to 2019). In addition, the elevation data for each state was acquired from 1980, with the understanding that the elevation of each state is not time sensitive and would not have changed drastically if the measurements were acquired in 2019. The year in which the data was acquired for each indicator variable in our framework is indicated in Table A2.
Furthermore, most of the data acquired are entered directly into the framework as raw values, meaning that they did not require additional computations before the LVI was calculated. However, some indicator variables under exposure, sensitivity, and adaptive capacity required further processing to be amenable and included in the analysis. Indicator variables under the exposure that required initial computations included annual average wind speed, humidity, annual precipitation, number of days with higher than 0.1 inches or more of precipitation, and annual temperature. The National Center for Environmental Information (NCEI) provides annual averages of each indicator for various weather observation stations located in each state. The values for every available weather observation station within each state were spatially averaged over the state and temporally averaged over the 1950 to 2018 period before being used in our LVI calculations.
The indicator variables requiring initial computation under sensitivity included the Palmer Drought Index (PDI) and the number of smokers. The National Oceanic and Atmospheric Administration (NOAA) collects monthly PDI values from weather observation stations throughout the US every year. The 2019 annual average was calculated for each station and then averaged amongst all the stations within a state. We also calculated the number of smokers using data from the United Health Foundation, which provided the percentages of smokers for every state. To accurately convey the proportions between the states, the state's population for that year was multiplied by its respective percentage of smokers. Finally, for adaptive capacity, only the indicator variable pertaining to the total area of lakes had to be computed. The original data provided the area for each indi- The indicator variables requiring initial computation under sensitivity included the Palmer Drought Index (PDI) and the number of smokers. The National Oceanic and Atmospheric Administration (NOAA) collects monthly PDI values from weather observation stations throughout the US every year. The 2019 annual average was calculated for each station and then averaged amongst all the stations within a state. We also calculated the number of smokers using data from the United Health Foundation, which provided the percentages of smokers for every state. To accurately convey the proportions between the states, the state's population for that year was multiplied by its respective percentage of smokers. Finally, for adaptive capacity, only the indicator variable pertaining to the total area of lakes had to be computed. The original data provided the area for each individual lake. Thus, we had to aggregate the area for all lakes to produce the cumulative lake area in each state.
The motivation for including the selected indicator variables in our framework was based on current risk assessment information suggested by the open literature, such as potential health risks due to wildfires [36]. Other examples include indicator variables pertaining to fuel, weather, and topography that are important drivers of wildfire danger and behavior, as referenced heavily in the literature [37,38]. Environmental indices such as the PDI and air quality were also included. While we acknowledge that there are many fire indices that could be integrated [14], we selected PDI because of its available spatial and temporal data for our study and because PDI is a useful indicator in describing an essential environmental factor (drought) required for the potential onset, ignition, and behavior of a wildfire [39]. Adding more fire indices and sub-indices would add redundancy to our framework.

LVI Calculation
Subsequently, we calculate the LVI and the corresponding contributing factors for each of the analyzed states based on our developed framework (Table A2). Our methods for computing the LVI follows a similar approach to [18,27]. Before the computation, we need to interpret whether the magnitude of each indicator value, under each contributing factor, is influencing the contributing factor positively or negatively. If the indicator variable is affecting the contributing factor negatively, then the inverse value is taken.
To compute LVI, we first compute the standardized index (SI) for each indicator variable, where I, is the original indicator variable for each individual state, Imax and Imin represent the state with the maximum and minimum value, respectively, corresponding to that particular indicator, we use Equation (1).
Second, the major component (MC) value for each state is computed by averaging the standard indices, over the number (n) of all indicators used in each major component, as in Equation (2).
Third, each contributing factor (CF) is computed by taking a weighted average of each computed major component. This is done by multiplying each major component by its number of indicators (Wi), as in Equation (3).
Finally, the LVI for each state is computed by combining the contributing factors of exposure (E), adaptive capacity (AC), and sensitivity (S), as in Equation (4).
The weighted balance function is applied to this method, as followed by [18]. The weighted function gives equal weighting to each indicator variable, despite how many indicators are present within the framework. This weighted approach is often used when determining vulnerability in data-scare regions. Once the LVI is computed for each state, a constant value of 0.5 is added to each LVI to simply aid in visualizing and interpreting the rank of LVI [26].

Validation Framework Approach
We subsequently applied a PCA to our indicator variables in order to gain confidence in the structure of our framework. PCA is a variable-reduction technique that takes a large set of variables and organizes them into a smaller set of principal components. For the purposes of this study, PCA was used to verify our framework by ensuring the indicator variables were loading into the "proper" major components that they were assigned. When conducting a PCA, four assumptions are made about the dataset, namely (1) the variables are measured at the continuous level, (2) there is a linear relationship between the variables, (3) there is adequate sample size, and (4) the dataset contains no outliers [40]. In addition, two tests are conducted to determine whether PCA is a suitable method for validating our framework: the Kaiser-Meyer-Olkin (KMO) sampling adequacy test [41] and Bartlett's Test of sphericity [42]. The KMO test measures the proportion of variance among the indicator variables that may be caused by underlying factors. KMO is an average of the measure of sample adequacy (MSA) for each indicator variable within their respective major component. MSA values range from 0 to 1 and represent the extent of a given indicator belonging to a group [43]. Smaller KMO values indicate fewer correlations between a given variable and the other indicators. Therefore, if the KMO value is less than 0.5, the results from a PCA will not be useful because the indicators do not share high correlations with each other. From Table A4 in the Appendix A, KMO values range mostly between 0.5 and 0.8, suggesting a strong sampling adequacy. Bartlett's test of sphericity is conducted to determine whether the correlation matrix of the indicators is an identity matrix. The null hypothesis is that the indicators are orthogonal (uncorrelated). For this study, if the indicator variables are uncorrelated, then they are unsuitable for this factor analysis. The values for this test range from 0 to 1, with 0 representing a rejection of the null hypothesis (meaning that the indicator variables are correlated). In addition, a significance value that is less than 0.05 indicates that PCA will provide helpful information. The values for the Bartlett's test of sphericity in our results mostly range from 0 to 0.3, suggesting that the variables chosen are correlated. Table A4 in the Appendix A provides the KMO and Bartlett test scores for each major component by using the indicator data gathered from the 14 states.
Once the indicator variables we selected had passed these tests, a PCA was conducted. The normalized data input for PCA were the standardized values for each indicator. The PCA gives insightful data such as a correlation matrix, communalities, and total variance explained. However, the output that helped reorganize and strengthen our framework was the component matrix. The component matrix displays the Pearson correlations between the indicator variables and principal components. The component matrix was used to verify whether the indicator variables loaded into their respective major components. This indicates that they are measuring the same underlying construct and are, therefore, correctly grouped accordingly in our framework.
Apart from the added confidence we gain from applying the PCA, we also conduct a sensitivity analysis to test whether our framework provides robust livelihood vulnerability results for each state. This is conducted by slightly perturbing our framework (through randomly selecting to omit one indicator variable at a time from a major component) and then re-running the LVI calculations. This will, thereby, provide twelve different framework scenarios with LVI results generated for each state. The original LVI results (current framework) is compared to the LVI output from each scenario to establish whether the top three LVI states and the lowest three LVI states remain consistent throughout 90% of the runs. If the LVI ranks remain the same (or are similar within reason) for more than 90% of the runs, these results provide additional validation to the framework. Refer to Table A8 for a synthesis of the scenarios. We acknowledge that other scenarios may be possible (by interchanging some indicator variables amongst the contributing factors) but this would lead to many other possible scenarios and has already been tested by the PCA.

LVI
We compute the LVI for each of the 14 American states analyzed ( Figure 3). Most of the states we analyzed exhibit similar LVI values. However, Arizona and New Mexico experience the greatest livelihood vulnerability, with an LVI of 0.57 and 0.55, respectively. In contrast, California, Florida, and Texas experience the least livelihood vulnerability to wildfires (0.44, 0.35, 0.33, respectively) ( Figure 4). To understand these LVI results, we delve into analyzing each contributing factor.

Exposure
First, we examine each state's susceptibility to wildfire by examining the exposure contributing factor. The exposure results indicate that California, Nevada, and Arizona exhibit the highest exposure to wildfires (0.63, 0.52 and 0.49, respectively) while Oklahoma, Florida, and Montana have the least exposure (0.25, 0.21 and 0.19, respectively) ( Figure 5a). To understand the exposure results, we assess the four major components of exposure (wildfire, topography, weather, and weather extreme events) for each state Figure 5b). Wildfires (blue) is predominant for California, Texas, and Arizona. This is because these states experience the highest number of wildfires and the largest acres burnt due to wildfires in 2019. Nevada and Arizona also experience relatively higher values of weather (yellow), which indicate favorable weather conditions for the development of wildfires, such as relatively higher wind speeds and lower humidity. In addition, weather extreme events (green) represent extreme wildfire and extreme heat events and are most prevalent in California and Nevada.

Exposure
First, we examine each state's susceptibility to wildfire by examining the exposure contributing factor. The exposure results indicate that California, Nevada, and Arizona

Exposure
First, we examine each state's susceptibility to wildfire by examining the exposure contributing factor. The exposure results indicate that California, Nevada, and Arizona  these states experience the highest number of wildfires and the largest acres burnt due to wildfires in 2019. Nevada and Arizona also experience relatively higher values of weather (yellow), which indicate favorable weather conditions for the development of wildfires, such as relatively higher wind speeds and lower humidity. In addition, weather extreme events (green) represent extreme wildfire and extreme heat events and are most prevalent in California and Nevada. The major component, topography, represents mean height and highest elevation for each state. Topography is important because higher elevations in complex terrain can be conducive to the propagation of wildfire behavior, add uncertainties to the prediction of the wildfire rate of spread [44], and make fire suppression efforts more challenging. Thus, states with higher topographic values could potentially be more at risk, or dangerously affected by wildfires. Nevada also ranks high in topography. While topography is also relatively high for other states, such as Wyoming and Utah, other major components, such as wildfires, weather, and weather extremes, are negligible, thereby reducing the overall exposure of wildfires in these states. Furthermore, Florida, Oklahoma, and Montana have the lowest exposures because all of their major components under exposure are ranked very low in comparison to the other states.

Sensitivity
Second, we assess the degree to which each state is affected by wildfires by investigating the sensitivity contributing factor. The results for sensitivity ( Figure 6a) show California as the most sensitive state to wildfires (0.84). This is followed by Texas, with a sensitivity of 0.66. Montana and Wyoming are the least sensitive. California, Texas, and Florida are the most sensitive to wildfires because they yield the highest values of each major component under sensitivity (demographic, ignition causes, and environmental index) (Figure 6b). Demographic comprises sub-components, such as the wildland-urban interface (WUI) and population. States with larger WUI areas or higher populations within a WUI, would be more sensitive to wildfires because they are within a region more The major component, topography, represents mean height and highest elevation for each state. Topography is important because higher elevations in complex terrain can be conducive to the propagation of wildfire behavior, add uncertainties to the prediction of the wildfire rate of spread [44], and make fire suppression efforts more challenging. Thus, states with higher topographic values could potentially be more at risk, or dangerously affected by wildfires. Nevada also ranks high in topography. While topography is also relatively high for other states, such as Wyoming and Utah, other major components, such as wildfires, weather, and weather extremes, are negligible, thereby reducing the overall exposure of wildfires in these states. Furthermore, Florida, Oklahoma, and Montana have the lowest exposures because all of their major components under exposure are ranked very low in comparison to the other states.

Sensitivity
Second, we assess the degree to which each state is affected by wildfires by investigating the sensitivity contributing factor. The results for sensitivity ( Figure 6a) show California as the most sensitive state to wildfires (0.84). This is followed by Texas, with a sensitivity of 0.66. Montana and Wyoming are the least sensitive. California, Texas, and Florida are the most sensitive to wildfires because they yield the highest values of each major component under sensitivity (demographic, ignition causes, and environmental index) (Figure 6b). Demographic comprises sub-components, such as the wildland-urban interface (WUI) and population. States with larger WUI areas or higher populations within a WUI, would be more sensitive to wildfires because they are within a region more exposed to wildfire events. Ignition causes attributed to outdoor activities, such as campfires and smoking, would also increase the potential inception of human-caused fires. In addition, states that experience poorer air quality and more drought will be more sensitive during and after wildfire events and seasons. The environmental index remains relatively constant among all states (yellow). However, California and Texas are the most sensitive states because they are driven primarily by the major components of ignition causes (red) and demographic (blue). The least sensitive state is Montana (0.08) because, in comparison to the other states, all its major components are ranked relatively low. addition, states that experience poorer air quality and more drought will be more sensitive during and after wildfire events and seasons. The environmental index remains relatively constant among all states (yellow). However, California and Texas are the most sensitive states because they are driven primarily by the major components of ignition causes (red) and demographic (blue). The least sensitive state is Montana (0.08) because, in comparison to the other states, all its major components are ranked relatively low.

Adaptive Capacity
Third, we assess the ability of each state to withstand or recover from wildfires by analyzing the contributing factor of adaptive capacity. Our results indicate that California, Texas, and Florida exhibit the greatest adaptive capacity to wildfires (0.69, 0.67 and 0.48, respectively) while Oregon, Idaho, and Montana are the least adaptive (0.15, 0.12, 0.12, respectively) (Figure 7a). The reasons for the adaptive capacity disparities among the states have to do with the major components (or capitals) that each state has (natural, physical, human, social network, and financial) Table A2.

Adaptive Capacity
Third, we assess the ability of each state to withstand or recover from wildfires by analyzing the contributing factor of adaptive capacity. Our results indicate that California, Texas, and Florida exhibit the greatest adaptive capacity to wildfires (0.69, 0.67 and 0.48, respectively) while Oregon, Idaho, and Montana are the least adaptive (0.15, 0.12, 0.12, respectively) (Figure 7a). The reasons for the adaptive capacity disparities among the states have to do with the major components (or capitals) that each state has (natural, physical, human, social network, and financial) Table A2. during and after wildfire events and seasons. The environmental index remains relatively constant among all states (yellow). However, California and Texas are the most sensitive states because they are driven primarily by the major components of ignition causes (red) and demographic (blue). The least sensitive state is Montana (0.08) because, in comparison to the other states, all its major components are ranked relatively low.

Adaptive Capacity
Third, we assess the ability of each state to withstand or recover from wildfires by analyzing the contributing factor of adaptive capacity. Our results indicate that California, Texas, and Florida exhibit the greatest adaptive capacity to wildfires (0.69, 0.67 and 0.48, respectively) while Oregon, Idaho, and Montana are the least adaptive (0.15, 0.12, 0.12, respectively) (Figure 7a). The reasons for the adaptive capacity disparities among the states have to do with the major components (or capitals) that each state has (natural, physical, human, social network, and financial) Table A2. What drives the adaptive capacity to be relatively high for California and, to a slightly lesser extent, Texas, are their social network (green) physical capital (red) and financial capital (orange) (Figure 7b). These two states have social structures in place to facilitate safety measures in times of wildfires such as allocating firefighters and first responders to wildland fire emergencies. These states are also more equipped with transportation accessibilities, such as closer airports and access to public roads, in case of major wildfires. California and Texas also have greater access to communication within their households, including internet signals for receiving warning alerts, both of which can be beneficial to one's livelihood during the state of an emergency wildfire evacuation. These states also rank highly in financial capital, such as having relatively higher household incomes and fire management assisted grants, which can lend financial support during wildland fire emergency hazards. Additionally, Florida also has a high adaptive capacity that is primarily driven by its natural capital. It has the largest water area of all the states analyzed, thereby providing the state with water resources for fire suppression.
In contrast to the states with the highest adaptive capacity, Montana, Idaho, and Oregon rank very low in all capitals. Moreover, while some states rank high in one major component, it suffers in others, thereby driving down the rank of its overall adaptive capacity value. For example, New Mexico has a relatively high human capital in comparison to other states, which corresponds to residential density and occupation; however, all its other capitals are negligible, resulting in an overall low adaptive capacity to wildfires. This emphasizes the need to evaluate all the contributing factors in adaptive capacity to obtain a holistic view of the allotted resources available to aid in wildfire resiliency measures. Adaptive capacity is one of the most important determining factors in risk assessment, as highlighted by [45]. who showed that wildfire hazard potential can be reduced once the adaptive capacity of the state is taken into consideration.  Table A4 in the Appendix A shows the results after running the KMO and Bartlett test. All of the values from the KMO test are at least 0.5, which is the minimum required value to conduct a PCA as described in [41]. The only major component that is not at least 0.5 is that of weather, which has a value of 0.488. Previous research such as [46] suggests a KMO value of at least 0.6 in order to proceed with PCA. However, due to the small sample size and indicators tested per PCA (adaptive capacity, 13; exposure, 11; sensitivity, 9) it is difficult to achieve a KMO value of at least 0.6. In addition, in this study, PCA was not utilized for its typical purpose of reducing variables, but rather, performed to verify whether the indicators within each major component loaded onto one principal component. Table A4 in the Appendix A also contains the results for the Bartlett test. Some of the major components achieved a desirable value of less than 0.05. However, some had values higher than 0.05. This is not an issue for two reasons. First, the major components that had a value greater than 0.05 had only two indicators to test. Only having two variables to create a correlation matrix would make it very difficult to achieve a value below 0.05. Second, the purpose of conducting a Bartlett test is to assess whether the correlation matrix diverges significantly from an identity matrix for data reduction [47]. Since the goal of the PCA is not variable reduction, the correlation matrix only needed to be proven as not being an identity matrix, that is, a value closer to 0 than 1.
After computing the PCA, we analyzed the generated component matrices. To validate the framework, the indicators had to have a strong loading into their respective major components. A strong loading is considered to be any value above 0.5 and suggests that the indicators are measuring the same underlying construct. Despite the fact that a PCA was conducted for each major component, the results are compiled into three tables (Tables A5-A7 in the Appendix A), one for each contributing factor. Overall, most of the indicators demonstrated a strong loading into their respective major components. However, there were some indicators that had weak loadings, under a value of 0.5, for example, annual average wind speed and annual average temperature in exposure. These indicators had a factor loading of 0.17 and 0.39, respectively, for the major component of weather. These low values indicate an inverse relationship between the other indicators under weather [48]. When a state is characterized by higher wind speed and temperature, they are more likely to be exposed to wildfires. The other indicators under weather involve humidity and precipitation. If a state is characterized by higher humidity and precipitation, then they are less likely to be exposed to wildfires in that same year. The same logic can be applied to the following indicators: acres of forests, number of timber/woodworkers, and annual PDI. These indicators all have negative loadings for their respective major components. These inverse relationships were reflected in the calculation of the LVI. With PCA verifying the structure of our framework, the validity of the LVI results is strengthened.

Sensitivity Analysis
Another test to validate our framework, included a sensitivity analysis. We perturbed the framework by removing one indicator variable at time and testing the rank of the new LVI results. A total of twelve scenarios were conducted (Table A8). Under exposures, the indicators removed were wildfire occurrence, mean height above sea level, wind speed, and extreme heat. Despite these exposure perturbations, the LVI rank for the top three (Arizona, New Mexico, Idaho) and lower three LVI states (Texas, Florida, California) remained the same as the original LVI outputs.
Similarly, the same analysis was carried out for sensitivity for which indicators: WUI area, number of campsites, and the air quality index were removed. The LVI results did not vary for these scenarios. Applying this approach to adaptive capacity, one indicator variable was removed at a time under each major component (area of lakes, miles of public roads, persons per household, number of firefighters, median household income). While the results remained the same for most perturbed scenarios of adaptive capacity, omission of the median household income changed the top LVI rank to include Colorado and not Idaho. The top LVI states were, Arizona, Colorado, followed by New Mexico. However, the lower three LVI states remained the same.
Overall, only one of the 12 scenarios showed a slight shift in the LVI rank for one of the states, but the top two (Arizona, and New Mexico) were still ranked accordingly. The results remained consistent for 92% of the runs. The results from this sensitivity analysis provides validation on applying our framework to quantify livelihood vulnerability.

Contribution of LVI
The main findings indicate that Texas, Florida, and California exhibit the lowest livelihood vulnerability to wildfires, while Idaho, New Mexico, and Arizona experience the greatest. Assessing each contributing factor and its respective major components and subcomponents have provided an in-depth analysis of why the livelihood vulnerability of some states to wildfires are higher than others. Many media and scientific reports constantly show California as the state with the most dangerous and destructive wildfires, especially in recent years. The NIFC report showed that California had the highest acres burned and maximum damages in 2018 among all the states. According to the 2019-2020 California Budget Summary [49], approximately ten of the most destructive wildfires in California have occurred since the year 2015. Thus, one might think that California, with the highest exposure, would have the highest LVI. Our study indicates that, while California is the most exposed, and sensitive to wildfires (Figure 8), it has a very high adaptive capacity to help offset its livelihood vulnerability. The California Administration has implemented solutions and recommendations to reduce wildfire risk to improve the state's emergency preparedness, response, and recovery capacity, and to further protect vulnerable communities. The 2019-2020 state budget includes 918 million dollars in additional funding to comply with these efforts [49]. For these reasons, it is evident why California exhibits a lower livelihood vulnerability to wildfires, relative to other states. capacity to help offset its livelihood vulnerability. The California Administration has implemented solutions and recommendations to reduce wildfire risk to improve the state's emergency preparedness, response, and recovery capacity, and to further protect vulnerable communities. The 2019-2020 state budget includes 918 million dollars in additional funding to comply with these efforts [49]. For these reasons, it is evident why California exhibits a lower livelihood vulnerability to wildfires, relative to other states. Similarly, Texas has the lowest LVI of all the states analyzed. Despite its high sensitivity, its exposure to wildfire is relatively lower than more than 25% of the other states and has the second-highest adaptive capacity. Texas is highly sensitive to wildfires. According to Texas A&M Forest Service (2020), there have been over 150,000 wildfires consuming more than 9 million acres since 2005 with 71,499 wildfires in 2017 alone [50]. Indeed, 90% of wildfires in Texas are human caused as a result of debris burning, sparks from welding and grinding equipment, poorly discarded smoking materials, vehicles' exhaust systems, and arson. Moreover, according to Headwater Economics (2018) parts of Texas that are experiencing the fastest population growth are spatially correlated with regions of highest wildfire threat and greater proportions of vulnerable people [51]. These Similarly, Texas has the lowest LVI of all the states analyzed. Despite its high sensitivity, its exposure to wildfire is relatively lower than more than 25% of the other states and has the second-highest adaptive capacity. Texas is highly sensitive to wildfires. According to Texas A&M Forest Service (2020), there have been over 150,000 wildfires consuming more than 9 million acres since 2005 with 71,499 wildfires in 2017 alone [50]. Indeed, 90% of wildfires in Texas are human caused as a result of debris burning, sparks from welding and grinding equipment, poorly discarded smoking materials, vehicles' exhaust systems, and arson. Moreover, according to Headwater Economics (2018) parts of Texas that are experiencing the fastest population growth are spatially correlated with regions of highest wildfire threat and greater proportions of vulnerable people [51]. These factors contribute to the sensitivity of Texas. However, we suggest that similar to California, Texas has a very high adaptive capacity, which drastically influences its livelihood vulnerability to wildfires. This high adaptive capacity is driven primarily by social network, physical capital, and financial capital. According to the Texas A&M Forest Service (2020), Texas has resources to deploy wildfire risk information and create awareness about wildfire concerns across the state through using a Texas Wildfire Risk Assessment Portal (TxWRAP) [50]. Furthermore, data produced from this portal is part of the Texas Wildfire Risk Assessment Project (WRA) that has further positioned the Texas Forest Service as a national leader in wildfire protection planning. These resources have positioned Texas to help withstand natural hazards related to wildfires.
Additional considerations should also be taken into account for states like Arizona that exhibit a high LVI. Arizona has high exposures of wildfires and high sensitivity to environmental indices such as drought and poor air quality. According to the Arizona Commerce Authority, Arizona is among the top three states with the highest rates of popu- lation growth in the nation [52]. There have been more than 120,000 new residents (doubled California's 50,635 new residents) in the 2018-2019 time period alone, with a projected population of over 10 million people by 2050 [52]. It can be assumed that with such growth, urbanization, transportation, and communication services will increase, thereby, making Arizona more sensitive to wildfire risk, as nine out of 10 wildland fires are started by humans according to the Arizona Department of Forestry and Fire Management [53].
We emphasize here that vulnerability is one factor in a risk assessment function, and thus, does not quantify the overall "risk". Therefore, the resultant LVI assessment is not meant to negate the fact that California, for example, is still at high risk of wildfire impacts, but instead, provides a relative measure that compares, specifically, the aspect of livelihood vulnerability to wildfires among the 14 states. To this end, it is important that we reiterate that despite a low LVI value in California, wildfires are continuing to worsen each year causing increased casualties, significant environmental impacts, and socio-economic damages. For example, the frequency of small (less than 500 acres) of wildfires have increased predominantly across western and central California and that wildfire season is lengthening to include the month of July [54]. There are also future concerns for the state of California, despite having a low LVI. We acknowledge and emphasize that the resultant exposure to wildfire in California is the highest amongst all states, thereby requiring continuous observations and monitoring. According to [55], the increased number of fires in California is due to a combination of climate change that has heightened hot and dry conditions and fire suppression policies that have allowed the accumulation of fuels in the landscape. As stated by numerous dependencies in the California Forest Carbon Plan in 2018, wildfire emissions are projected to increase by 19%-101% using the 1961-1990 years as the baseline period [56].
If global greenhouse gas emissions continue to increase at its current rate, wildfire smoke will increase, only exacerbating these emissions and worsening the current health impacts. Therefore, looking to the future, mitigation and resiliency strategies need to be developed and adopted for the high livelihood vulnerable states, such as Arizona. In addition, continued efforts are required for relatively low LVI states that have a high exposure such as California in order to facilitate and provide resources to help mitigate wildfire hazards in the future.
The need to adopt contemporary practices is beneficial for resiliency and mitigation methods. For example, proposed policies focusing on fire suppression and prevention became prevalent in the early 1900s and represented the foundation of California's economic theory of wildfire management, following a massive fire that had burned 3 million acres in Montana, Idaho, and Washington [57][58][59].However, according to the recent California Policy Center (2017), fire suppression techniques only worked as short-term solutions, resulting in over 100 million dead or dying trees, overgrown forests, and fuel accumulation, increasing the risk for dangerous wildland fires [60]. Thus, the continued need for evolving and enhancing fire management techniques and practices is essential for accurately monitoring and improving wildfire risk assessments.

Conclusions
Across the US, wildfires can produce catastrophic environmental and socio-economic impacts. To quantify these risks across multidimensional, socio-economic, and biophysical variables, we produce a framework to compute a livelihood vulnerability index for the top 14 American states that are most at risk for wildfires. Our framework comprises contributing factors (exposure, sensitivity, and adaptive capacity), major components, subcomponents, and indicator variables. Our framework was further justified by performing a principal component analysis to provide additional confidence in our approach.
Our results indicate that the states of Arizona and New Mexico experience the greatest livelihood vulnerability, with an LVI of 0.57 and 0.55, respectively and California, Florida, and Texas experiencing the least livelihood vulnerability to wildfires (0.44, 0.35, 0.33, respectively). LVI is weighted strongly on the contributing factors. For example, while California has a high exposure and sensitivity to wildfires, it has high adaptive capacity capitals that offset these factors. Additionally, livelihood vulnerability depends largely on sensitivity indicator variables, such as population density. We acknowledge that, with Arizona's high LVI and steady population growth, continued wildfire risk management and urban planning strategies are essential for reducing the biophysical and socio-economic impact of wildfires in the future and to further avoid an increase in its LVI.
This study provides a first order approximation that uses secondary (available census) data to provide a novel quantitative account of livelihood vulnerability through multidimensional factors. While we acknowledge that higher resolution scale (perhaps at the community level) would provide additional valuable information, this requires primary data acquisition through field surveys and interviews at the local scale, which is beyond the scope and feasibility of this work. However, the results generated from our relatively coarser scale analysis, can point towards high order evaluations that can be conducted at finer scales in future studies.
The results from this study are critical to researchers, government, and policymakers in identifying, allotting, and providing better resiliency and adaptation measures to support the states that are most vulnerable to wildfires. Further research can be conducted, following the same framework for each of the state's geo-political subdivisions in order to better understand the risk and vulnerability of growing wildland-urban interface zones and to determine what urban boundary limitations should be considered for risk assessment studies. Moreover, additional research can be conducted to assess future LVI scenarios by employing high-resolution forecast models to help guide future wildland fire exposure projections in vulnerable communities within the US.

Data Availability Statement:
The data presented in this study are available within the article and Table A2 in the Appendix A.
Appendix A Table A1. LVI terminology definitions, color coordinated by major components in each contributing factor: adaptive capacity (green), exposure (red) and sensitivity (blue). Gray highlights denote terms that are frequently used in livelihood vulnerability literature.

Contributing factor
Overarching biophysical and socio-economic factors used to calculate LVI (exposure, adaptive capacity, and sensitivity) [18] Adaptive capacity The system's (state's) ability to adjust to a perturbation or disturbance and cope with consequences [61] Exposure The degree, time and or extent a system (state) is in contact with, or subject to a perturbation (e.g., wildfire events) [61] Sensitivity The degree to which a system (state) is modified or affected by the perturbation or set of disturbances [61] Major component The first level of divisions within each contributing factor [18] Financial capital Considers financial resources a system (state) has in order to help adapt to an exposure (wildfire) e.g., grants, income [62] Human capital Considers human resources and level of education and productive skills of people in a system (state) e.g., occupation type [62] Natural capital Considers natural resources in a system (state) that helps a system adapt to an exposure (wildfire) e.g., lakes, forests [62] Physical capital Considers materials and resources that a system (state) has to help adapt to an exposure (wildfire) e.g., transportations and communication types, infrastructure and livestock [62] Social network Considers social constructs that are in place by a system (state) in order to help adapt to an exposure (wildfire) e.g., safety practices, clubs, networks, affiliations [62] Wildfire Occurrence Metric used to quantify the number of wildland fires in a state, e.g., wildfire occurrence, loss of wildland Topography Considers metrics used to quantify topography of landscape, e.g., elevation height Weather Considers the meteorological metrics that influences wildfire behavior, e.g., air temperature Weather Extreme Events Considers metrics that quantifies extreme environmental conditions conducive for wildfires e.g., extreme heat Demographic Considers metrics that describes the population structure of a state, e.g., population density Ignition causes Considers metrics pertaining to potential ignition sources for the onset of a wildfire, e.g., smoking Environment Indices Indices that compute a potential risk related to wildfires, e.g., an air quality index Subcomponent The second level of divisions within each major component [18] Indicator variables Measurable units of data for each sub-component

Livelihood vulnerability index (LVI)
A vulnerability assessment tool to address issues of sensitivity, exposure and adaptive capacity to climate change (wildfire) in fire-prone communities [18] Fire 2021, 4, 54 Table A2. LVI framework outlining the indicator variables, sub-components, and major components that feed into each contributing factor (exposure, adaptive capacity, and sensitivity). A rationale for each indicator is provided along with a description as to how the indicator is interpreted into the LVI framework. Asterisk (*) denotes indicator variables for which the inverse values were applied in the LVI equation. In addition, the data source and the year of analysis is presented.

•
Forests provide an abundance of fuel, which is required for wildfires to start and spread [66] • While states with large forested acreage can be an asset, it also provides larger amounts of fuel to burn, spawning the development of large wildfires, therefore, the inverse value of "acres of forests" are considered and can be interpreted as, states with "less" acres of forest, may have a higher capacity to manage and mitigate wildfires risk