Wildland fire smoke adds to disproportionate PM_2.5 exposure in the United States

Abstract

Wildland fire (i.e., prescribed fire and wildfire) smoke exposure is an emerging public health threat, in part due to climate change. Previous research has demonstrated disparities in ambient fine particulate matter (PM_2.5) exposure, with Black people, among others, exposed to higher concentrations; yet it remains unclear how wildland fire smoke may contribute to additional disproportionate exposure. Here, we investigate the additional PM_2.5 burden contributed by wildland fire smoke in the contiguous United States by race and ethnicity, urbanicity, median household income, and language spoken at home, using modeled total, non-fire, and fire PM_2.5 concentrations from 2007 to 2018. Wildland fires contributed 7% to 14% of total population weighted PM_2.5 concentrations annually, while non-fire PM_2.5 concentrations declined by 24% over the study period. Wildland fires contributed to greater PM_2.5 exposure for Black and American Indian or Alaska Native people, and those who live in non-urban areas. Disproportionate mean non-fire PM_2.5 concentrations for Black people (9.1 μg/m³, compared to 8.7 μg/m³ overall) were estimated to be further exacerbated by additional disproportionate concentrations from fires (1.0 μg/m³_, compared to 0.9 μg/m³ overall). These results can inform equitable strategies by public health agencies and air quality managers to reduce smoke exposure in the United States.

Graphical Abstract

1. Introduction

Due to the successful regulation of ambient sources of air pollution through the Clean Air Act, there has been substantial improvement in United States (US) air quality over the last three decades, including a decline in fine particulate matter (PM_2.5, particulate matter with an aerodynamic diameter less than or equal to 2.5 μm).¹ However, in some areas of the US, recent increases in wildfire activity are contributing to flattening or reversing of the decreasing PM_2.5 trends.^2–4 PM_2.5 is a primary health concern because it is a major component of the complex mixture of pollutants in smoke, and exposure can lead to numerous negative health outcomes, including respiratory- and cardiovascular-related effects as well as early death.⁵ Recent studies focusing on wildfire smoke demonstrate similar associations with health effects and are consistent with the findings for ambient PM_2.5 exposure (i.e., exposure during a typical day).^6–8 Indeed, national studies examining the public health burden of wildland fire smoke have identified widespread impacts, with estimates of thousands of attributable hospital admissions and deaths leading to billions of dollars of economic costs.^{9, 10} Moreover, PM_2.5 concentrations are likely to increase in the future as climate change amplifies wildfire conditions, resulting in the emission of more smoke,^11–13 and as the use of prescribed fire increases to reduce fuel loads following more than 100 years of fire suppression.^{14, 15}

Given the anticipated increase in wildfire activity, it is increasingly important to understand the sociodemographic disparities in wildland fire smoke exposure in relation to ambient non-fire PM_2.5 concentrations, both to better communicate the risks posed by smoke and to inform equitable strategies to protect public health. Sociodemographic disparities in ambient PM_2.5 exposure have been well documented, with people of color, including Black people in particular, experiencing a larger burden of pollution than other racial and ethnic groups nationally.^{5, 16–18} Studies of wildland fire smoke exposure examining sociodemographic disparities in the US are much more limited, but suggest that smoke exposure may be more prevalent among some sociodemographic groups. Fann et al. (2018)¹⁰ showed that a higher percentage of Black and Native American people lived in areas with >75^th percentile wildfire PM_2.5 compared to Asian and white people from 2008 to 2012. Vargo et al. (2023)¹⁹ reported similar findings in a national study from 2011 to 2021, finding that high density smoke disproportionately affected people of color and those with limited English language capacity. In contrast, Burke et al. (2021)² reported higher wildfire-specific PM_2.5 concentrations among non-Hispanic white populations.

While the aforementioned studies report seemingly contrasting findings that point to the need for further investigation, they were also limited to focusing only on wildfire-specific PM_2.5 exposure and did not account for underlying disparities in PM_2.5 exposure from non-fire sources that may also exist. The previous studies that examined disparities in smoke exposure used different methods to assess PM_2.5 concentrations across different overlapping time periods and they focused on fire-attributed PM_2.5 as an indicator of smoke. Further, these studies analyzed exposure by racial and ethnic groups using various methods of classification and focused on different groups.^{2, 10, 19} Here, we seek to address these limitations by investigating exposure to PM_2.5 from wildland fire smoke and from non-fire sources simultaneously, while also considering language spoken at home, income, and urbanicity. Specifically, we use modeled PM_2.5 exposure in the contiguous US (CONUS) from 2007 through 2018, from wildland fire and non-fire sources, to address the following questions:

1. How do wildland fires contribute to total US PM_2.5 concentrations and how is this changing over time?

2. How does wildland fire smoke contribute to disparities in total PM_2.5 exposure among different sociodemographic groups?

The results of this study can help identify population groups disproportionately impacted by both wildland fire and non-wildland fire PM_2.5 to help support targeted public health interventions.

2. Materials and methods

2.1. PM_2.5 concentrations

PM_2.5 concentrations assigned to Census tracts were used as a surrogate for population-level PM_2.5 exposure. Daily 24-hour mean PM_2.5 concentration estimates for 2007 to 2018 were obtained from the US Environmental Protection Agency (US EPA)’s Fused Air Quality Surface Using Downscaling (FAQSD) data product.^20–23 This product statistically combines Community Multiscale Air Quality (CMAQ) 12 km resolution gridded model outputs with monitoring observations from US EPA National Air Monitoring Stations/State and Local Air Monitoring Stations using a Bayesian space-time downscaler approach to estimate PM_2.5 concentrations at the centroid of each Census tract. To estimate PM_2.5 concentrations attributable to wildland fire, we used previously published CMAQ simulations run with and without wildland fire emissions inputs over the years 2007 to 2018 at 12 km resolution.^{24, 25} Wildland fire emissions were day- and location-specific based on multiple types of data sources: satellite fire detections, incidence reports, and fire perimeter databases. Fuel type, consumption, and emissions were estimated with the BlueSky v3.5.1 modeling framework.^{26, 27} Fire emissions using this approach and CMAQ modeling system have been shown to compare well with surface level PM_2.5 and field study measurements.^28–31 Uncertainty exists in all aspects of fire emissions estimation, but large fires are best represented since those typically have multiple types of information available about location, size, and timing. Annual mean PM_2.5 concentrations were calculated for the “total” and “non-fire” datasets, and “fire” PM_2.5 was taken as the simple difference of total PM_2.5 minus non-fire PM_2.5. The resulting annual mean concentrations were assigned to 2010 US Census tract centers of population³² using cubic spline interpolation.^33–35 This attributional modeling was used to estimate the percent of PM_2.5 concentrations from fires, which was then multiplied by the annual mean PM_2.5 concentration in the FAQSD dataset for each Census tract to calculate the wildland-fire attributed annual mean PM_2.5 concentration (i.e., fire PM_2.5). A similar approach has been used by other studies^36–38 to combine fire-attribution modeling with statistically fused estimates of total PM_2.5. The resulting distributions of total, fire, and non-fire PM_2.5 concentrations are hereafter referred to as the “CMAQ” dataset. Separately, to characterize the high extreme of exposure for population subgroups, the population weighted 98^th percentiles of daily total PM_2.5 concentrations from FAQSD were calculated for each year.

Since air quality monitors cannot differentiate between PM_2.5 from wildland fire and non-fire sources, models are often relied upon to estimate wildland fire contributions to total PM_2.5 concentrations. However, models estimating PM_2.5 concentrations during wildfire smoke events have varying levels of agreement and poor correlation with ground monitors at high concentrations.³⁹ To address these concerns, we conducted a sensitivity analysis to compare the results of our approach with an entirely different method of calculating wildfire-specific PM_2.5 concentrations using data from Childs et al. (2022),⁴⁰ hereafter referred to as the Childs Wildfire Exposure Estimate (CWEE). In that work, daily wildfire PM_2.5 concentrations were predicted at Census tracts using a machine learning model with satellite, Lagrangian modeling, and ground monitor inputs. This dataset notably focuses specifically on wildfire and likely excludes smoke from smaller prescribed fires, which differs from the CMAQ dataset used in the main analysis that includes both prescribed fire and wildfire (wildland fire = wildfire + prescribed fire). Results obtained using the CWEE were compared with results using the CMAQ method.

2.2. Sociodemographic variables

All sociodemographic information was captured at the 2010 Census tract scale. Total residential population count and race and ethnicity were from the 2010 Census,⁴¹ median household income was from 2008 to 2012 American Community Survey (ACS) sampling,⁴² and language spoken at home was from 2006 to 2010 ACS 5-year estimates.⁴³ Although some studies of air pollution exposure disparities have used Census block groups (subsets of Census tracts), analyses of Census tract resolution data have been shown to be highly correlated with results from analyses using finer spatial resolutions.⁴⁴ Regional analyses were performed using the CONUS regions defined by the US National Climate Assessment.⁴⁵

The population residing in each Census tract was categorized by race and ethnicity into the following groups: Hispanic, Non-Hispanic (NH) White (White), NH Black (Black), NH American Indian and Alaska Native (AIAN), and NH Asian (Asian), including 97.8% of the total population. Individuals self-identifying in more than one group were excluded. In addition, language spoken at home was examined to investigate exposure among non-English speakers, categorized into three groups for people living in each tract: 1) only English spoken at home; 2) language other than English spoken at home, speaks English “very well”; and 3) language other than English spoken at home, speaks English less than “very well”. These simplified groups were chosen to provide a high-level overview of Census data detailing many individual languages. To investigate potential socioeconomic factors potentially contributing to disparities in exposure while accounting for regional differences in income and cost of living, median household income for each Census tract was ranked into quartiles according to the following protocol, based on a modification of that used in Hirsch and Schinasi (2019):⁴⁶ 1) for Census tracts within Core-Based Statistical Areas (CBSAs), Census tracts were ranked within their respective CBSA; 2) for Census tracts outside of CBSAs, Census tracts were ranked within their respective state. To capture disparities in PM_2.5 exposure by urbanicity, Rural-Urban Commuting Area (RUCA) classifications⁴⁷ were used to describe urbanicity based on primary traffic flows between Census tracts, yielding more detail than the urban and rural dichotomous classification provided by the Census. The ten primary RUCA codes were simplified into five categories using the previous approach of Messer et al. (2010):⁴⁸ urban core (RUCA code 1), suburban (RUCA code 2), micropolitan (RUCA codes 3, 4, 5, and 6), small town (RUCA codes 7, 8, and 9), and rural (RUCA code 10). Each sociodemographic variable investigated was available for >99% of CONUS Census tracts.

2.3. Population exposure to PM_2.5

The distributions of PM_2.5 across all Census tracts on each day were aggregated into annual mean and overall mean (2007 to 2018) measures of exposure. Correlation between Census tract-level annual means of each of the datasets was examined using Pearson’s r. To examine exposure for each population subgroup, absolute PM_2.5 burden was calculated overall and for demographic subgroups using the population weighted mean (PWM) (equation 1),

$$PWM=\frac{\sum _{i=1}^n{P}_i{C}_i}{\sum _{i=1}^n{P}_i}$$ (1)

where $P_i$ is the population residing in Census tract $i$ and $C_i$ is the annual mean PM_2.5 concentration assigned to the Census tract. Proportional PM_2.5 burden is calculated as the simple ratio of the subgroup exposure over the overall exposure (equation 2).

$$Proportionalburden=\frac{{PWM}_{subgroup}}{{PWM}_{overall}}$$ (2)

A proportional burden higher than one indicates an exposure higher than the total national exposure and a proportional burden less than one indicates a lower exposure. PWM_overall was set to the CONUS PWM for both CONUS and regional analyses. All concentrations reported in this paper are population weighted unless otherwise noted. Additionally, to characterize the high extreme of daily PM_2.5 exposure, population weighted 98^th percentiles were calculated using the open-source Statsmodels library stats.weightstats.DescrStatsW function.⁴⁹ All analyses were performed using open-source Python 3.11⁵⁰ software.

3. Results

3.1. Fire and non-fire PM_2.5 concentrations and trends

During the 2007 to 2018 study period, non-fire population weighted PM_2.5 concentrations decreased by 24%, while fire PM_2.5 concentrations varied between a minimum of 7% in 2009 and a maximum of 14% of total PM_2.5 in 2017 (Figure 1; Figure S1). Decreases in total and non-fire PM_2.5 over the study period were most prominent in the East and Midwest (Figure S2; Figure S3), while regions with elevated fire PM_2.5 were more variable year-to-year (Figure S4). Southeast fire PM_2.5 concentrations were higher in 2007 and 2008 than in other years of the study period. The western US also experienced elevated fire PM_2.5 in 2007 to 2008, then a notable increase again in 2015, 2017 and 2018 (Figure S4). Evaluating trends at the Census tract level, areas with above-median annual mean concentrations for both fire and non-fire PM_2.5 remained fairly consistent throughout the study period despite decreasing total PM_2.5 concentrations in most of the country, with areas of California and the Midwest above the median for both in all years. In later years, more Census tracts in the inland Northwest and Southeast had above-median concentrations for each (Figure S5). Thus, the spatial distribution of fire PM_2.5 was more variable than that of non-fire PM_2.5 over the years studied.

Figure 1.

Yearly overall population weighted mean non-fire and fire PM_2.5 concentrations in the contiguous United States from 2007 to 2018.

In the primary analyses, we considered the full distribution of concentrations because disparities in PM_2.5 exposure may exist even at low concentrations. To additionally show where fires contributed to relatively high PM_2.5 concentrations within the CONUS, we accounted for Census tracts where annual concentrations exceeded the overall (2007 to 2018) national population weighted total PM_2.5 mean of 9.6 μg/m³ in individual years (Figure 2). The number of Census tracts exceeding 9.6 μg/m³ decreased in each year from 2007 through 2018, although areas in the West, Midwest, and South had many Census tracts greater than 9.6 μg/m³ in 2018. As non-fire concentrations of PM_2.5 have decreased,^{1, 51} the proportion of tracts above 9.6 μg/m³ due to the contributions of fires has increased (Figure 2; Figure 3).

Figure 2.

Contiguous US Census tracts with annual mean PM_2.5 concentrations exceeding 9.6 μg/m³, considering total PM_2.5 and PM_2.5 from non-fire sources from 2007 to 2018. Light blue denotes Census tracts that would not be above 9.6 μg/m³ if not for contributions from wildland fires.

Figure 3.

Number of people living in contiguous US Census tracts with annual means > 9.6 μg/m³ PM_2.5 (i.e., the overall 2007 to 2018 population weighted mean), considering total PM_2.5 and PM_2.5 from non-fire sources from 2007 to 2018.

3.2. Disparities in wildland fire and non-fire PM_2.5 exposure

Averaged from 2007 to 2018, wildland fires are estimated to have contributed 0.9 μg/m³ (9.2%) to a total population weighted mean PM_2.5 concentration of 9.6 μg/m³ (Table 1). Fire and non-fire PM_2.5 concentrations were weakly anticorrelated (Pearson’s r: −0.19; Table S1), suggesting that regions with high fire PM_2.5 often do not often coincide with regions with high non-fire PM_2.5. Furthermore, the burden of fire PM_2.5 was not distributed equally among the population (Table 1; Figure 4). Among racial and ethnic groups, PM_2.5 concentrations from wildland fires were highest for AIAN people, receiving 1.1 μg/m³ (12.8% of total PM_2.5). Asian people were exposed to the least wildland fire smoke, with 0.8 μg/m³ (7.5%) of PM_2.5 concentrations attributable to fires. By region, the Northwest and Southeast had the highest concentrations of fire PM_2.5 (1.2 and 1.4 μg/m³; 15.4 and 14.5% of total PM_2.5), and the Midwest and Northeast experienced the least (0.6 and 0.4 μg/m³; 6.1 and 3.9% of total PM_2.5). Those living in urban Census tracts experienced a smaller fire PM_2.5 burden (0.8 μg/m³; 7.9% of total PM_2.5) than those living in less dense communities (1.1 to 1.2 μg/m³; 12.2 to 14.0% of total PM_2.5). Smaller concentration differences were observed by language spoken at home and quartile of median household income (Table 1; Figure 4).

Table 1.

Population weighted annual mean PM_2.5 concentrations (2007 to 2018) in the contiguous US attributed to all sources (total), non-wildland fire (non-fire), wildland fire (fire), and the Childs et al. (2022)⁴⁰ estimate of wildfire PM_2.5 (CWEE), by National Climate Assessment (NCA) regions, primary Rural Urban Commuting Area codes (RUCA), racial and ethnic groups, language spoken at home, and quartiles of median household income.¹

		CMAQ Total (μg/m3)	CMAQ Non-fire (μg/m3)	CMAQ Fire (μg/m3)	CMAQ % Fire	CWEE (μg/m3)
Overall		9.6	8.7	0.9	9.2	0.4
NCA Region	Midwest	10.4	9.8	0.6	6.1	0.5
	Northeast	9.4	9.0	0.4	3.9	0.3
	Northern Great Plains	7.4	6.3	1.1	14.5	0.7
	Northwest	7.9	6.7	1.2	15.4	0.7
	Southeast	9.4	8.0	1.4	14.5	0.3
	Southern Great Plains	9.6	8.6	1.0	10.1	0.4
	Southwest	9.7	8.8	0.9	9.5	0.4
RUCA	Urban core	9.8	9.0	0.8	7.9	0.4
	Suburban	9.2	8.1	1.1	12.2	0.4
	Micropolitan	9.0	7.8	1.2	13.1	0.4
	Small town	8.8	7.5	1.2	14.0	0.4
	Rural	8.2	7.1	1.1	14.0	0.5
Race and ethnicity	AIAN	8.2	7.2	1.1	12.8	0.4
	Asian	10.1	9.3	0.8	7.5	0.4
	Black	10.0	9.1	1.0	9.6	0.4
	Hispanic	9.8	9.0	0.8	7.9	0.3
	White	9.4	8.5	0.9	9.5	0.4
Language spoken at home	Only English	9.6	8.7	0.9	9.3	0.4
	Other than English, does not speak English well	9.6	8.8	0.8	8.8	0.4
	Other than English, speaks English well	9.6	8.7	0.8	8.7	0.4
Income quartile	1 (lowest)	9.7	8.8	0.9	8.9	0.4
	2	9.6	8.7	0.9	9.2	0.4
	3	9.5	8.6	0.9	9.3	0.4
	4 (highest)	9.6	8.7	0.9	9.2	0.4

Figure 4.

Proportional burden of total (CMAQ Total), wildland fire (CMAQ Fire), non-fire (CMAQ Non-fire) PM_2.5 concentrations by Rural-Urban Commuting Area (RUCA) urbanicity classifications, race and ethnicity, language spoken at home, and median household income quartile.

These results for fire PM_2.5 contrast with the exposure for non-fire PM_2.5 for some population groups, but not others. Among racial and ethnic groups, non-fire PM_2.5 concentrations were estimated to be highest for Asian people (9.3 μg/m³) but fire PM_2.5 concentrations were lowest (0.8 μg/m³; 7.5% of total PM_2.5; Table 1). A similar pattern existed for Hispanic people, and to a lesser extent, for white people. For these groups, fire PM_2.5 added to the total PM_2.5 exposure but attenuated the proportional burden of total PM_2.5 compared to non-fire PM_2.5 concentrations (Figure 4). In comparison to the overall population, Black people were exposed to higher concentrations of both fire PM_2.5 (1.0 μg/m³) and non-fire PM_2.5 (9.1 μg/m³), and as a result, total PM_2.5 proportional burden was higher compared to that of non-fire PM_2.5 (Figure 4). That is, we estimate that the non-fire PM_2.5 exposure disparity for Black people is exacerbated by exposure to fire PM_2.5.

Comparing the characteristics of Census tracts with greater than the population weighted mean fire PM_2.5 (0.88 μg/m³) or non-fire PM_2.5 (8.7 μg/m³) concentrations to those with mean concentrations below that value, AIAN people comprised a larger percent of the population in areas with high fire PM_2.5 (0.9%) compared to areas with high non-fire PM_2.5 (0.3%; Table S2). People living in areas with both high fire PM_2.5 and non-fire PM_2.5 are concentrated in the Southeast, where 48.7% of the population lives in areas with both high non-fire PM_2.5 and fire PM_2.5, whereas 22.3% or less of the population lives in areas that experience both in other regions (Table S2; Figure 5). Areas with both high non-fire PM_2.5 and fire PM_2.5 had fewer white people (59% of the population, compared to 64% overall) and more Black, Hispanic, and Asian people than the overall population. These areas had higher income, as well (Table S2; Figure 5).

Figure 5.

Locations of Census tracts with greater than the annual mean population weighted mean concentrations from 2007 to 2018 of non-fire PM_2.5 (8.7 μg/m³)_, fire PM_2.5 (0.88 μg/m³), and both fire and non-fire PM_2.5 greater than mean concentrations.

Across all sociodemographic characteristics investigated, proportional burden of fire PM_2.5 varied substantially more by year than the proportional burden of non-fire PM_2.5 (Figure 6). Black or AIAN people were the racial and ethnic groups most exposed to fire PM_2.5 in each year studied, but temporal trends in the two groups’ exposure varied. Black people were exposed to the highest fire PM_2.5 concentrations in 2007 to 2008, when Southeast fire PM_2.5 concentrations were high, and the least concentrations in 2016 to 2018 (Figures 6, S6, and S7). The final three years studied coincided with increased western US fire PM_2.5 concentrations (Figure S6) that likely drove higher exposure among Asian and Hispanic people (Figures 6 and S7). Asian or Black people experienced the highest non-fire PM_2.5 concentrations in each year and AIAN people consistently experienced the least non-fire PM_2.5 concentrations. The relative burden of both fire PM_2.5 and non-fire PM_2.5 by income quartile remained similar in each year (Figure S6). Figures S8 to S10 show trends in annual mean concentrations by urbanicity, language spoken at home, and median household income quartile. Although more variable than proportional burden of non-fire PM_2.5, the rank of urbanicity categories by fire PM_2.5 concentrations changed little year to year. People who speak a language other than English at home were exposed to higher concentrations compared to English speakers in 2017 and 2018, whereas concentrations were approximately equal to or lower than English speakers in earlier years. Among sociodemographic characteristics analyzed, the rank order of fire PM_2.5 concentrations varied less year-to-year by urbanicity than it did among racial and ethnic groups and by language spoken at home.

Figure 6.

Annual proportional burden of PM_2.5 concentrations in 2007 to 2018 by Rural-Urban Commuting Area (RUCA) urbanicity classifications, race and ethnicity, language spoken at home, and median household income quartile for fire PM_2.5 (CMAQ Fire) and non-fire PM_2.5 (CMAQ Non-fire).

The 98^th percentile of daily total PM_2.5 concentrations, investigated as a metric of peak concentrations that is often due to wildfire, declined from 31 μg/m³ in 2007 to a minimum of 18 μg/m³ in 2016, with higher concentrations in 2017 and 2018 (20 and 22 μg/m³; Figure S11). Among racial and ethnic groups, Asian people live in places with higher 98^th percentile total PM_2.5 concentrations than each of the other racial and ethnic groups considered for each year in 2007 to 2018 (Figure S12). The 98^th percentile of concentrations in 2017 and 2018 increased steeply in the western US, among Asian and Hispanic people, those who primarily speak a language other than English at home, and those living in urban areas (Figure S12). This increase coincided with the increase in area burned in California in those years⁵² and a large increase in fire PM_2.5 was simultaneously observed in the CWEE. In combination with the trends in annual mean fire PM_2.5 (Figure S6), these results indicate a shift in the demographics of the burden of fire PM_2.5 in 2017 and 2018 compared to earlier years studied.

3.3. Disparities in exposure within regions of the CONUS

The proportional burden of total, fire, and non-fire PM_2.5 within regions was investigated with a focus on the Southeast and Northwest (Figure 7) because those regions had the highest fire PM_2.5 concentrations (Table 1). Proportional burden in other regions is shown in Figures S13 to S17. Considering racial and ethnic groups, the burden of fire PM_2.5 varied more by region than that of non-fire PM_2.5. Fire PM_2.5 concentrations were higher for AIAN people compared to other groups in all regions except those with the lowest fire PM_2.5 concentrations (i.e., the Midwest, Northeast, and Southwest). Patterns of fire PM_2.5 burden among other racial and ethnic groups varied across regions, with Black people, for example, among the most burdened in the Southeast but among the least burdened in the Northwest. Black and Asian people are among the groups living in areas with the highest non-fire PM_2.5 burden across all regions. The proportional burden of fire PM_2.5 by language spoken at home is similarly varied by region. Disparities by urbanicity are largely consistent across regions, with lower fire PM_2.5 and higher non-fire PM_2.5 concentrations in urban areas. Proportional burden of each fire and non-fire PM_2.5 was similar across income quartiles in each region. Across regions of the CONUS generally, the proportional burden of fire PM_2.5 varied more consistently by urbanicity than by racial and ethnic group or language spoken at home.

Figure 7.

Proportional burden of total (CMAQ Total), fire (CMAQ Fire), non-fire (CMAQ Non-fire) PM_2.5 concentrations by Rural-Urban Commuting Area (RUCA) urbanicity classifications, race and ethnicity, language spoken at home, and median household income quartile in the Northwest and Southeast.²

3.4. Sensitivity analysis

Census-tract level annual mean fire PM_2.5 modeled using our CMAQ approach is moderately correlated with predictions from CWEE (Pearson’s r: 0.49; Table S1). The CWEE overall population weighted mean is 0.38 μg/m³, which is considerably lower than the 0.88 μg/m³ predicted using CMAQ (Table 1). Comparing the main analysis results with those produced in Childs et al. (2022)⁴⁰ (i.e., CWEE), some results are robust regardless of the choice of fire PM_2.5 modeling method, including higher fire PM_2.5 burden among AIAN people compared to the overall population (proportional burden of 1.2 and 1.1 using CMAQ and CWEE PM_2.5, respectively). However, results are less consistent for other groups. Black people are predicted to have less exposure than the overall population using the CWEE PM_2.5 predictions (CWEE proportional burden: 0.96; CMAQ fire proportional burden: 1.1; Figure S18). Regionally, CWEE PM_2.5 is highest in the northern great plains (proportional burden: 1.9), whereas CMAQ fire PM_2.5 is highest in the southeast (proportional burden: 1.6). The CWEE approach does not emphasize prescribed fire, which is highest in the Southeast⁵³ and included in the CMAQ approach. Trends also vary in specific years between the two approaches; CWEE PM_2.5 predicts much higher relative increases in population weighted mean concentrations across all racial and ethnic groups in 2011 and 2018 in particular (Figure S6). The results using the two models of fire PM_2.5 diverge in 2018, where the CWEE PM_2.5 increased for all racial and ethnic groups relative to 2017 and CMAQ fire PM_2.5 decreased. Western US burned area increased between 2017 and 2018⁵⁴ and the two estimates of fire PM_2.5 disagree on whether this led to increased population weighted fire PM_2.5 concentrations (Figure S6).

4. Discussion

We report that wildland fires contribute to relatively more PM_2.5 exposure for Black and AIAN people, and those who live in non-urban areas. Furthermore, the contributions of wildland fires may exacerbate previously identified total PM_2.5 disparities for Black people, who tend to live in areas where both fire and non-fire PM_2.5 concentrations are high in comparison to concentrations experienced by the overall population. Our findings vary by region, with the Northwest and Southeast experiencing the highest fire PM_2.5, but these results differed in a sensitivity analysis using another dataset of fire PM_2.5. Previous studies have investigated racial and ethnic disparities in wildland fire smoke exposure at the national level, but none comprehensively included the major racial and ethnic groups classified by the US Census, nor put those disparities in context of non-fire PM_2.5 disparities.

Our findings generally align with other literature investigating racial and ethnic disparities in fire PM_2.5 exposure, and we note differences in observed results arising from how fire PM_2.5 is modeled. Our observation of higher fire PM_2.5 concentrations for Black people is consistent with Fann et al. (2018)¹⁰ for the CONUS and Johnson Gaither et al. (2019)⁵⁵ in Georgia. However, we did not observe higher exposure for Black people in the sensitivity analysis using the CWEE PM_2.5 dataset, likely attributable to the lack of focus on prescribed fire in Childs et al. (2022),⁴⁰ discussed further below. Prescribed fire is an important source of wildland fire smoke in the Southeast, where more Black people live relative to other regions. Wildland fire in the southeastern US has been dominated by prescribed fire, and in contrast, the West has had more wildfires.⁵³ In addition, our results indicate a higher burden of fire PM_2.5 for AIAN people that has consistently been identified in the limited number of studies that considered their exposure.^{10, 56} Studies of wildfire PM_2.5 in California have identified higher exposure among AIAN people relative to the overall population,^{56, 57} and another identified elevated PM_2.5 exposures among Black and Hispanic populations in Los Angeles and San Francisco Bay during wildfire events in 2020,⁵⁸ although these are not directly comparable to our national study. Burke et al. (2021)² reported a positive association between wildfire PM_2.5 and percent white population, which is directionally consistent with our results, but they did not consider other racial and ethnic groups for which we identified a greater burden (i.e., Black and AIAN people). Vargo et al. (2023)¹⁹ identified a higher number of days with intense smoke in areas with a higher proportion of minority populations. We did not identify any previous studies of national-level fire PM_2.5 exposure that have investigated disparities by income, language, or urbanicity. Our analysis ^{10, 19, 55}characterizes disparities arising from the regional distributions of the population and wildland fire smoke. Systematic neighborhood-scale disparities in burdens of pollution, such as those demonstrated for traditional anthropogenic sources of PM_2.5,¹⁶ are unlikely to exist for wildland fire smoke due to the variable location of wildland fires and geographic scale of the smoke plumes (i.e., more local for prescribed fire and regional for wildfire).^{59, 60}

Public health interventions generally focus on measures that individuals and communities can employ to reduce smoke exposure.⁷ Unlike stationary or mobile sources of air pollution, wildfire smoke cannot be controlled through emission control technologies, and prescribed fire, a major tool used to reduce the risk of catastrophic wildfire, also emits smoke. Instead, public health protection measures currently employed include clean air shelters, buildings with adequate heating, ventilation, and air conditioning systems where those without access to filtered air can shelter during extreme smoke events; personal protective equipment such as high-efficiency respirators (e.g., N95 masks) used outdoors during smoke events; and portable air cleaners for improving indoor air quality during smoke events.⁶¹ The use of these interventions may vary among sociodemographic groups due to inconsistent health risk messaging, lack of communication regarding the health risks of smoke, or a variety of factors that may affect individuals’ ability to take action to reduce their exposure.⁶¹ For example, those living in the central valley of California experience high concentrations of fire PM_2.5 (Figure 5), and many residing in this location are Hispanic and employed as outdoor workers,^{62, 63} which could contribute to challenges around communicating the health risks of smoke and actions to reduce exposure if bilingual materials are not presented. In addition, outdoor workers may also have challenges in reducing exposure due to their occupation. Therefore, while the regional and national scale results of this study can help inform where disparities in exposure to both fire and non-fire PM_2.5 occur, more detailed analyses may be necessary at individual locations to help support local and more targeted interventions as the vulnerable populations in each location may comprise different demographic characteristics.

Using available PM_2.5 monitoring and modeling methods presents challenges for understanding the spatiotemporal distribution of fire PM_2.5. Because the regulatory air quality monitoring network focuses primarily on large population centers and was not designed for wildland fire smoke, exposures to both non-fire and fire PM_2.5 are less understood in rural areas.^{7, 64} Although we rely on models to predict PM_2.5 concentrations continuously throughout US Census tracts, those located farther from monitoring sites have larger uncertainties associated with the predictions.^20–23 Additionally, because the fraction of PM_2.5 attributed to fires is modeled and applied to a separate dataset of total PM_2.5 concentrations, it is unclear how well the estimates of fire PM_2.5 analyzed in this study reflect true fire PM_2.5 concentrations. Model predictions of PM_2.5 have been shown to be less reliable during intense wildfire smoke events.³⁹ A study using CMAQ to investigate fire PM_2.5 showed that the model underpredicts concentrations when monitors observe high concentrations and overpredicts concentrations when monitors observe low concentrations during wildland fire smoke events.⁶⁵ Furthermore, the Census tract-scale comparisons are limited by the coarser 12 km CMAQ grid cells. In a sensitivity analysis, we compared the results obtained using the CMAQ approach with those obtained using a recently published daily fire PM_2.5 dataset,⁴⁰ although their approach is also subject to limitations. Childs et al. (2022)⁴⁰ relies heavily on the NOAA Hazard Mapping System to predict wildfire PM_2.5, which tends to miss small fires, many of which are prescribed fires, and best characterizes smoke events with regional impact on air quality.⁵⁹ Further evaluation and improvement of models of fire PM_2.5 may increase understanding of patterns of exposure and improve research of the short and long-term health effects of smoke exposure.

Beyond limitations of the PM_2.5 data, the sources of demographic information used in this analysis may affect the results. Among other factors, we did not consider demographic estimates for years other than 2010, and thus, it is possible that demographic changes within the study period may alter the reported PM_2.5 concentration estimates for subgroups. Furthermore, the racial and ethnic categories tracked by the Census are coarse, people who selected more than one individual racial group were not included, and some people (e.g., people of Middle Eastern or North African descent) may not identify as part of any of the officially listed groups.⁶⁶ Lastly, our approach using residential Census tract concentrations as a surrogate for exposure does not consider how behavioral patterns may vary by sociodemographic group and affect exposure to PM_2.5 from fires and other sources.

Overall, we estimated that fires comprise 9.2% of total annual population weighted mean PM_2.5 concentrations in the United States from 2007 to 2018, and the burden of fire PM_2.5 is distributed among the population differently and more variably than non-fire PM_2.5. This is unsurprising given the stochasticity of fire and the smoke emitted, yet it is notable, given that it likely requires different public health outreach and interventions compared to those traditionally employed for air pollution through regulatory actions. Among racial and ethnic groups, disparities in non-fire PM_2.5 concentrations for Black people may be exacerbated by the contributions of fires. AIAN people live in areas with the lowest non-fire PM_2.5 but the highest fire PM_2.5, and Asian people live in areas with the highest non-fire PM_2.5 but lowest fire PM_2.5. Those living outside of urban areas experience a higher proportion of their total PM_2.5 burden from wildfire than those living in urban areas. As concentrations of non-fire PM_2.5 are reduced through the successful implementation of the Clean Air Act¹ and the burden of wildfires is increasing,^{2, 3, 11–13} investigating the distribution of fire PM_2.5 is increasingly important to identify communities disproportionately burdened so that interventions can be targeted to equitably reduce smoke exposures.

Synopsis.

This study investigates the contribution of wildland fire smoke to total fine particulate matter exposure for different population groups to inform equitable exposure reduction strategies in the United States.