The distributional incidence of wildfire hazard in the western United States

In recent years, the western US has faced dramatic increases in wildfire activity and damages attributed to wildfires (Abatzoglou and Williams 2016, Westerling 2016, Wang et al 2020, Buechi et al 2021). This trend is a result of rising wildfire activity and—in the case of wildfire damages to homes and private property—growth in the number of properties in harm's way. Due to climate change (Westerling et al 2006, Abatzoglou and Williams 2016, Goss et al 2020) and the accumulation of heavy fuel loads in many western forests (Keane et al 2002, Naficy et al 2010), area burned in large forest fires in the western US increased by approximately 1200% from the 1970s to 2012 (Westerling 2016). At the same time, the number of homes exposed to potential wildfire damage has grown. Between 1990 and 2010, the number of US homes in wildland-urban interface areas—those areas where development and wildland vegetation intermingle and homes are more likely to encounter wildfire danger—grew by over 40% (Radeloff et al 2018), while total wildland-urban interface area increased by more than 30%.

The rapid escalation of wildfire activity and associated damage to structures has resulted in a growing list of policy challenges, including increasing spending on wildfire management by federal and state agencies, burgeoning liability costs for electric utilities, and challenges to affordability and availability of homeowners' insurance in high wildfire hazard areas. At issue in addressing these challenges is who should bear costs associated with rising risk; should increasing costs of wildfire risk be borne primarily by those living in high wildfire hazard areas or should they be distributed more broadly? For many, the answer to this question will hinge on fairness concerns. Nevertheless, despite its relevance to a variety of wildfire-related policy questions, the distributional incidence of wildfire hazard is not well-understood.

This paper seeks to fill this gap. Using a rich property-level dataset of residential property values within the western US, Census economic and demographic data, and spatial wildfire hazard data, we contrast the distributions of property values and demographics across high and low wildfire hazard areas, and across areas that have and have not experienced recent wildfires. We also examine how the distributional incidence of wildfire hazard varies geographically across the western US. Our results provide important context for a variety of policy debates related to rising wildfire risk, and contribute to an understanding of the environmental justice implications of increasing hazards under climate change.

This analysis contributes to a large literature documenting differences in exposure to environmental hazards across socioeconomic groups. The literature studying differences in exposure to anthropogenic hazards such as pollution and waste facilities spans decades, generally finding that the burden of pollution is disproportionately borne by socially vulnerable groups (for reviews, see Ringquist 2005, Mohai et al 2009, Banzhaf et al 2019). In contrast, research describing differences in natural disaster risk faced by various demographic and economic groups has arisen more recently, in the wake of Hurricane Katrina (e.g. Walker and Burningham 2011, Grineski et al 2015, Montgomery and Chakraborty 2015, Bakkensen and Ma 2020). A broad finding of this emerging literature is that exposure to flood risk can differ among racial/ethnic groups and by economic status, but the relationship is complicated by the correlation between flood risk and valued coastal amenities (e.g. views or access to beaches); whereas exposure to anthropogenic hazards is primarily determined by household choices over where to live. In some coastal areas, demand may be driven more by preferences for amenities than by aversion to flood risk.

As in flood risk areas, correlation between wildfire hazard and local amenities complicates expectations regarding the distributional incidence of wildfire hazard (Stetler et al 2010). High wildfire hazard areas may be dominated by wealthy households who have chosen to live in high hazard forested areas with significant wildland amenities. Alternatively, high fire hazard areas may comprise largely low income or otherwise vulnerable households, who have chosen to live in a rural or outlying location due to affordability concerns. Thus, the distributional incidence of wildfire hazard is an empirical question. Compared to flooding though, relatively few studies have explored how wildfire hazard is distributed across economic and demographic groups.

Studies of environmental justice dimensions of wildfire risk exposure vary based on whether they consider differences in hazard or vulnerability across groups. Disaster risk is defined as potential for future losses from disaster events (Field et al 2012), and is the product of hazard (the likelihood of a destructive event), exposure (the presence of values at risk that could be potentially harmed), and vulnerability (the susceptibility of values at risk to damage or loss). Some studies have attempted to identify factors that affect vulnerability to fire, including parcel-level characteristics (e.g. building materials) and demographics (Paveglio et al 2016, Davies et al 2018, Palaiologou et al 2019). However, these studies have yielded mixed evidence on the relationship between demographics and vulnerability to wildfires, in part due to the difficulty of measuring vulnerability.

Other studies have attempted to identify areas of intersection between high social vulnerability—the socially-determined vulnerability to harm measured based on Census variables assembled in a 'social vulnerability' index (see Cutter et al 2003)—and high wildfire hazard (Lynn and Gerlitz 2005, Poudyal et al 2012, Wigtil et al 2016). While these studies are interested in social 'vulnerability', they differ from the studies mentioned above in that they focus on identifying how hazard is distributed across groups with varying levels of social vulnerability, rather than on identifying how vulnerability is distributed. In this way they are closely related to our analysis here. In the most comprehensive analysis of this type thus far, Wigtil et al (2016) found that high fire hazard locations tend to have lower social vulnerability on average, but within high fire hazard areas approximately 10% of the population is characterized by high social vulnerability.

We build on this analysis by using property-level data on locations and property values, allowing us to relate property values to wildfire hazard at the point location of each property and to perimeters of historical wildfires. We then analyze distributional incidence within the entire US West region, within individual states, and within counties. These two features of our analysis are important since they allow us to address concerns related to the 'Ecol. fallacy,' under which the use of more aggregate data may mask relationships between demographics and environmental characteristics at finer spatial scales (Banzhaf et al 2019). For example, relationships between population characteristics and wildfire hazard may depend on where within a block group people actually live relative to wildfire hazard (see figure A.1 available online at stacks.iop.org/ERL/17/064031/mmedia). Moreover, relationships between population characteristics and hazard at national scales may result more from broad-scale relationships between these variables (e.g. wildfire hazard and property value are both high in California). Whether this broad scale variation or finer scale variation is most relevant may vary depending on researchers' interests and motivations.

Recent studies have also examined the distribution of benefits from wildfire management and wildfire risk mitigation activities. With respect to fuels management projects, this research finds that agencies are more responsive to post-fire demands for projects by higher income communities (Anderson et al 2020) and that EJ communities were rarely consulted in project planning processes (Adams and Charnley 2020). With respect to wildfire suppression effort and expenditures, this research indicates that on a per-home basis lower-value homes receive a greater implicit fire suppression subsidy since they tend to occur in lower density areas where wildfire suppression expenditures are split between fewer homes (Baylis and Boomhower 2019), but that fires are more likely to stop spreading as they approach higher value homes (Plantinga et al 2020). By comparing the distribution of property values in high fire hazard areas to the distribution of property values among properties actually affected by wildfires between 2011 and 2018, we offer insights into the degree to which wildfire management has influenced the distribution of exposure to wildfires over this period.

3.1. Data

3.2. Regression analysis

Using the merged property and hazard data, we estimate a series of binned OLS regressions specified according to:

$$\begin{align} hwhp_{i} = \gamma_{d^s(i)} + \mu_{s(i)} + \varepsilon_{i}, \end{align} \tag{ 1 }$$

where d indexes the property value decile of property i within geography s, hwhp_i is an indicator variable denoting whether the property is in a high WHP area, and the fifth decile is in all cases the omitted category against which other deciles are compared. In regressions in which s indexes counties, for example, deciles are calculated within individual counties. In this case, $\mu_{s(i)}$ represents the share of high wildfire hazard properties among those in the fifth property value decile within county s. Parameters $\gamma_{d^s(i)}$ indicate how the share of high wildfire hazard properties departs from $\mu_{s(i)}$, on average, in other deciles. This specification allows us to consider how relative wildfire risk varies across the distribution of property values, where the relevant property value distribution can vary depending on the geography of interest.

4.1. Property values and demographic characteristics within high wildfire hazard areas

High fire hazard block groups have higher per capita income, a lower percentage of residents below the poverty line, a higher percentage of white residents, and a lower percentage of black and Hispanic residents (table 1). The differences are substantial, and persist whether we define high hazard block groups to be those in which more than 20% of properties have wildfire hazard above the 90th, 95th, or 99th percentile among all properties in the western US ⁵ . Per capita income is at least 15% higher in high wildfire hazard areas, and the share of white residents in these areas is at least 15 percentage points greater. While these characteristics are consistent with lower social vulnerability in high wildfire hazard areas, high fire hazard areas also have a higher share of Native American residents and a higher share of residents older than 65, indicating the presence of subpopulations in high fire hazard areas that may be higher in vulnerability. In the supplementary data (table A.1), we explore within-state differences in average demographic characteristics across high and low wildfire hazard areas.

Table 1. Differences in demographics across high and low wildfire hazard block groups.

	Low hazard	High hazard (90th perc.)	High hazard (95th perc.)	High hazard (99th perc.)
Per capita income	30 857	35 896	35 622	36 457
Percent below poverty line	15.4	12.0	11.8	11.6
Percent white (non-Hispanic)	63.7	78.2	80.4	84.8
Percent black	5.2	2.3	2.1	1.3
Percent Hispanic	30.1	18.6	17.1	13.9
Percent Native American	1.2	2.0	2.1	2.2
Percent over 65	11.2	14.0	14.7	16.3

Note: Columns 3–5 report mean values for blocks in which at least 20% of the properties are in areas where WHP is above the 90th, 95th, and 99th percentiles, respectively, among all homes in our data set. Low hazard block groups are those in which fewer than 20% of properties have WHP above the 90th percentile.

High fire hazard properties are also higher in value on average than low fire hazard properties. Figure 1 provides a kernel density plot (panel (A)) and heat map (panel (B)) illustrating how the distribution of property values varies with wildfire hazard. High fire hazard properties are defined in panel (A)—and for the remainder of the paper—as those with fire hazard above the 95th percentile among properties in the western US, while low hazard properties are those in which WHP equals zero. Across the western US, median property values are nearly $17 thousand greater in high fire hazard areas than in low fire hazard areas. Figure A.4 repeats this plot separately for each state. The relative distributions of property values vary substantially by state, with the difference between median properties in high and low hazard areas being more than $50 thousand throughout much of the interior western US (Colorado, Montana, Nevada, New Mexico, Utah, and Wyoming).

Zoom In Zoom Out Reset image size

However, while high hazard properties are more likely to be high in value, panel (B) shows that properties in the highest wildfire hazard deciles are also more likely than low wildfire hazard properties to be in the lowest property value deciles. This pattern is repeated individually in most western states, as indicated by figures A.5 and A.6 provides a figure analogous to figure 1 but for per capita income. Consistent with table 1 and figure 1, per capita incomes are greater in high fire hazard areas. In contrast to figure 1, figure A.6 indicates a disproportionately low share of the population in the lowest income decile live in high fire hazard areas.

We also find that properties in the upper end of the property value distribution are more likely to live in high wildfire hazard areas (figure 2). Regardless of the geography across which produce property value distributions, we find that properties in the seventh property value decile or higher are more likely to be in high wildfire hazard areas than low value properties. Properties in the tenth decile of property value by county are 3% points more likely than homes in the fifth decile to be in high wildfire hazard areas. Within the fifth decile of property value, approximately 4% of homes are in high wildfire hazard areas. Therefore our estimates imply that homes with value in the tenth decile for properties in their same county are more than 70% more likely to be in high hazard areas than homes in the fifth decile.

Zoom In Zoom Out Reset image size

The distribution of wildfire hazard depends somewhat on the set of households chosen as a comparison group. In figure 2, we observe the strongest relationship between high property value and wildfire hazard when we compare properties within the same county. This may be because within-county comparisons mute the effects of proximity to major urban centers and other confounding factors whose variation is primarily across counties. We also find that low property value homes are no more likely than median homes to be in high fire hazard areas when low property values are defined only based on within-county comparisons. This may likewise be because the relationship between low property values and high fire hazard at the state and regional level is driven by underlying differences across counties (such as distance from urban centers) and correlations with fire hazard.

4.2. Home value and historical wildfire occurrences

To assess how historical wildfire occurrences have been distributed across western properties, figure 3 compares the distribution of property values for properties facing high wildfire hazard to the distribution of 2010 property values among homes within historical wildfire perimeters. If wildfire occurrences were randomly distributed with respect to property values, we would expect to observe 10% of the properties inside a wildfire perimeter to come from each decile. On the other hand, if wildfire management provided substantially greater benefits to households living in higher value homes, we might expect to see a relatively lower percentage of high value homes within wildfire perimeters than within high fire hazard areas. However, across the western US as a whole, we find that historical wildfires disproportionately affected homes in the highest property value deciles (deciles 9 and 10) between 2011 and 2018.

Zoom In Zoom Out Reset image size

These results reinforce findings from the previous section that higher value homes face disproportionate wildfire risk; however, two qualifications should be noted. First, we measure property location relative to the fire perimeter, not whether the property was damaged or destroyed. While it is unlikely that all of the properties we identify as having been inside a fire perimeter were damaged or destroyed by fire, presence inside the fire perimeter nevertheless indicates increased risk of damage. Second, these results are highly influenced by a small number of fires. Between 2011 and 2018, nearly half (46%) of properties within fire perimeters were within the Camp and Tubbs fire perimeters. The figure shows how values of affected properties were distributed within these fires. The Tubbs Fire primarily impacted very high value properties. The Camp Fire primarily affected low or moderate value properties, though the impacts were distributed relatively uniformly. Overall, the degree to which wildfire impacts were concentrated among high value homes between 2011 and 2018 was moderated by the Camp Fire, due to its significant size and the area affected.

4.3. Geographic variation in the distributional incidence of wildfire hazard

Figure 4 illustrates how property values are distributed geographically within high wildfire hazard areas across the western US. As expected, the coast of California is dominated by high property value high wildfire risk areas. In California, Arizona, the Salt Lake City area, and the Colorado Front Range, there are cells with relatively high numbers of high hazard properties. However, across the remainder of the western US, high wildfire hazard areas are largely sparsely populated.

Zoom In Zoom Out Reset image size

The upper bar chart on the right indicates that large portions of high wildfire hazard areas in the western US consist mainly of relatively low value homes, while much smaller portions of high wildfire hazard areas contain primarily high value homes. To reconcile this result with the results in section 4.1—that homes in high wildfire hazard areas are higher value on average—the lower bar chart illustrates how number of properties per grid cell varies across property value deciles. High fire hazard areas consisting mainly of low value homes cover a much broader portion of the western US than those with higher value homes; however, these areas contain fewer high fire hazard properties than do areas where high fire hazard properties are primarily higher value because they are lower density on average.

This paper finds that those living high wildfire hazard areas are disproportionately likely to live in relatively high value homes, to have higher incomes, and to be white—demographic characteristics associated with low social vulnerability. As in high amenity yet high hazard coastal areas, it is likely that spatially-correlated wildland amenities (e.g. tree cover, views, and access to recreational opportunities) make properties in high wildfire hazard areas desirable for higher income households. Our findings suggest that, overall, households in wildfire hazard areas are more likely to be pulled to live in these areas by these amenities, rather than to be pushed to live in outlying wildfire hazard areas by affordability concerns. This finding is similar to those from the environmental justice literature on exposure to flood hazards, which finds high amenity coastal areas may be relatively high income, sometimes in spite of high hazard.

It is important to note that this paper does not identify, nor does it set out to identify, the effect of wildfire hazard on home prices; rather, we use home prices as a proxy for household wealth in order to understand how wildfire hazard is distributed across the population. Since 60% of the average middle class household's wealth is in their home, property value is a reasonable proxy for household wealth; however, we also test the relationships between wildfire hazard and other demographic variables measured by the US Census. The broad message of our analysis remains the same whether we examine property values or Census demographics.

A secondary finding of this paper is that high wildfire hazard areas are highly heterogeneous. While median property values in high wildfire hazard areas are higher than those in low wildfire hazard areas, there is significant overlap in the distribution of property values across the two types of areas (figure 1), and properties with very low values are somewhat over-represented in high wildfire hazard areas when compared to other properties in the western US or compared to properties in the same state (figure 2). As well, some demographic groups that may be more vulnerable, including very low income households, Native Americans, and the elderly, are disproportionately represented in high hazard areas. Members of these groups may be less able to adapt and recover from wildfire events than lower vulnerability households due to household wealth, insurance coverage, access to credit, or health. Though exposure to wildfire hazard falls disproportionately on higher income households with higher property values, vulnerability—susceptibility to damage or loss in the case of a wildfire event—may be distributed differently.

Finally, while properties in high wildfire hazard areas have higher values on average, the spatial distribution of those properties has implications for management. We find that because higher property value high wildfire hazard areas tend to be denser, most high wildfire hazard areas across the western US comprise relatively low value properties (figure 4). Since wildfire management is a local public good that benefits all households in a given area (Baylis and Boomhower 2019, Wibbenmeyer et al 2019), investments in wildfire management may be more efficient and more substantial in wealthier and more densely populated areas. WHP does not take into account geographic differences in wildfire management activities; however, historical fire perimeters may in part be a function of management choices. We found that between 2011 and 2018, there were not large broad-scale differences between the distribution of wildfire hazard across households and the distribution of wildfire occurrences. This finding suggests that differences in exposure to wildfire hazard are the primary driver of the distributional incidence of wildfire damages, and that any differences in benefits from wildfire management are not sufficient to have shifted the overall burden onto poorer communities. We note, however, that this finding is highly influenced by relatively few significant wildfire incidents, and additional research is needed to examine distribution of management effort directly and within individual wildfire incidents.

With respect to environmental justice, our findings indicate that, overall, exposure to wildfire hazard differs from exposure to anthropogenic hazards such as pollution or waste facilities, which disproportionately affect vulnerable communities. In California and increasingly in other parts of the western US as well, policymakers are weighing options for how to distribute the costs of wildfire risk across electric utilities, insurance companies, state and federal agencies, and households in both low and high hazard areas. The results of this paper suggest that many households affected by escalating wildfire risk may be capable of bearing increased costs associated with living in these high amenity locations. On average, policies that shift resources from low hazard areas to high hazard areas, provide subsidies to wealthier households, potentially raising fairness concerns. Nevertheless, high wildfire hazard areas are heterogeneous, and we find some more socially vulnerable populations (e.g. very low income households, Native Americans, the elderly) are disproportionately represented in high wildfire hazard areas. Therefore, addressing environmental justice concerns associated with costs of increasing wildfire hazard may call for a more geographically targeted approach focused on reducing the burden for the most vulnerable communities.

This work was supported by Wildfire Risks in Rural Communities: Measuring Impacts, Modeling Choices, Evaluating Policies (2021-67023-34483), a Grant from the US Department of Agriculture National Institute of Food and Agriculture. Data are provided by Zillow through the Zillow Transaction and Assessment Dataset (ZTRAX). More information on accessing the data can be found at www.zillow.com/ztrax. The results and opinions are those of the author(s) and do not reflect the position of Zillow Group.

The data generated and/or analysed during the current study are not publicly available for legal/ethical reasons. Please contact the the corresponding author for information about accessing the data.