Ambient air quality and health impacts of PM_2.5 from US residential wood combustion

Abstract

Residential wood combustion (RWC) is a primary heating fuel in just 2% of US homes. However, the 2023 release of the US Environmental Protection Agency’s National Emissions Inventory (NEI) found that RWC contributes ~28% of total wintertime fine particulate matter (PM_2.5) emissions, suggesting that ambient PM_2.5 concentrations could be substantially reduced if RWC were curtailed. Despite its contribution to PM_2.5 emissions, an assessment of the air quality, health, and distributional impacts of RWC using the updated NEI has not been performed. Here, we use a high-resolution (4 kilometers) air quality model and the updated NEI to evaluate wintertime RWC impacts over the contiguous United States. We find that RWC contributes 2.43 micrograms per cubic meter (21.9%) of winter population-weighted mean PM_2.5 concentrations, leading to ~8600 (confidence interval: 6500 to 9600) premature deaths annually. Moreover, nonwhite communities are disproportionately affected by RWC-related PM_2.5 and associated mortality, especially in urban areas. We suggest that policies targeting RWC could substantially reduce air pollution, improve health, and address distributional disparities.

Air pollution from residential wood burning contributes to thousands of premature deaths annually, with unexpected urban impacts.

INTRODUCTION

Air pollution is a prodigious global hazard, posing environmental, air quality, and health risks while contributing to millions of premature deaths each year. Exposure to fine particulate matter (PM_2.5), one of the United States (US) Environmental Protection Agency’s (EPA’s) criteria pollutants, has been extensively and robustly linked to adverse health outcomes, including respiratory and cardiovascular diseases, as well as increased mortality (1). The 2025 State of Global Air Report estimates that ambient outdoor PM_2.5 causes ~4.9 million deaths per year globally (2), more than any other category of pollution. In the US, PM_2.5 exposure is estimated to cause ~95,000 to 300,000 premature deaths per year (3), with exposure unevenly distributed across demographic and socioeconomic groups (4). Tessum et al. (5) found that non-Hispanic white populations are exposed to 17% less PM_2.5 pollution than is caused by emissions associated with their consumption and activities, whereas Black and Hispanic populations experience 56% and 63% higher PM_2.5 exposure than their respective emission contributions.

The combustion of wood for residential heating has been identified as a major source of both indoor and outdoor PM_2.5 globally (6–8). In the US, a range of residential wood combustion (RWC) appliances are used for space heating, cooking, and recreational purposes (9). Central RWC appliances, such as hydronic heaters and wood-burning furnaces, are used almost exclusively as primary heaters (i.e., heat sources that provide most or all of a household’s space heating needs). Wood stoves (~27% primary) and pellet-fired wood stoves (36% primary) are often used as secondary heat sources (i.e., sources that are used intermittently to offset costs or provide additional warmth alongside other systems, such as natural gas furnaces and electric heaters) (10). Conversely, fireplaces and fire pits are typically used for cooking and recreational burning rather than space heating (fig. S1) (11).

In this study, we focus exclusively on the ambient, or outdoor, component of RWC-derived PM_2.5. RWC is a potentially attractive target for ambient PM_2.5 mitigation in the US due to its relatively high proportion of emissions per energy production. Approximately 2% of homes use RWC as their primary source of heat, with only ~8% using RWC as a secondary heat source (12). However, RWC is estimated to produce ~485,000 tons of primary PM_2.5 annually (table S1), more than double the total estimated primary PM_2.5 emissions of the transportation sector (not considering road dust), and ~28% of total wintertime US PM_2.5 emissions (13). Though RWC produces higher PM_2.5 emissions per unit energy generated than other forms of residential energy/heating technologies, global RWC activity may be on the rise. Biomass burning has been presented as a natural, renewable carbon fuel source (14), leading many governments to implement policies incentivizing RWC as an alternative to fossil fuels, and ultimately contributing to recent increases in global biomass burning (15). In many regions, residential biomass burning also persists due to the widespread availability of wood and other solid fuels, as well as long-standing cultural and traditional practices surrounding their use (16). In the US, the relative importance of RWC has grown as emissions from other major sources, such as vehicles and power plants, have declined (17, 18).

An increase in global RWC, in particular, poses a public health risk because studies have shown that long-term exposure to PM_2.5 derived from woodsmoke is associated with equal or more toxic risk factors than PM_2.5 from non-woodsmoke sources (19–26). In the US, polycyclic organic matter from RWC is estimated to contribute ~25% of area source air toxic cancer risks (27) and woodsmoke PM_2.5 has also been associated with increases in short-term mortality [1.92%; 95% confidence interval (CI): 1.19 to 5.03 per 10 μg/m³ increase in PM_2.5] (28), making increases in wood-burning activity detrimental to both short- and long-term health via acute and chronic exposure. In addition to the adverse health impacts of RWC, RWC-derived air pollution exacerbates anthropogenic climate change through its role in climate forcing via the release of radiation-absorbing carbonaceous aerosols (29), highlighting a co-beneficial motivation for targeted mitigation.

Previous studies conducted in various global regions have assessed the impacts of RWC emissions on air quality and public health. Many non-US studies have used air pollution models, including high-resolution chemical transport models (CTMs), to estimate the pollutant and health impacts of RWC PM_2.5 (30–34). For the US, we identified just three contiguous United States (CONUS)–focused CTM studies that have investigated questions pertinent to the impacts of RWC. These three studies used different CTMs [i.e., Community Multiscale Air Quality (CMAQ) and Comprehensive Air Quality Model with Extensions (CAMx)] at different spatial resolutions (i.e., 12 and 36 km), but each relied on 2005 National Emissions Inventory (NEI) estimates. Fann et al. (35) estimated ~8000 premature CONUS deaths per year from RWC, Caiazzo et al. (36) estimated ~41,800 premature CONUS deaths per year from residential and commercial combustion combined, and, most recently, Penn et al. (37) estimated 10,000 annual premature CONUS deaths associated with primary RWC. In addition, we found one Southeast US–focused study that attempted to assess RWC PM_2.5 impacts using the Integrated Source Apportionment Method (CMAQ-ISAM) and the 2008 NEI; however, their modeled concentrations severely underestimated observations from a source apportionment network (38). While the relative consistency of past CONUS estimates lends confidence to our understanding of RWC impacts, reliance on now-dated emissions estimates and use of somewhat coarse CTM simulations suggest state-of-the-science RWC impact estimates are needed.

To update and improve CONUS RWC impact estimates, we leverage the 2020 NEI (released in 2023) and a CONUS-wide, high-resolution (4 km) CTM simulation. RWC emissions estimates in the 2020 NEI used an updated methodology that leveraged novel survey data and updated spatial processing techniques (13). Using these updated RWC emissions, we perform 4-km CONUS-wide simulations with the two-way coupled Weather Research and Forecasting (WRF)–CMAQ modeling system to estimate the contribution of RWC to PM_2.5 concentrations. We then use our WRF-CMAQ simulations to provide state-of-the-science estimates of exposure, health, and distributional impacts of RWC-derived PM_2.5 pollution.

RESULTS

RWC PM_2.5 concentrations

RWC contributes to winter PM_2.5 concentrations throughout the CONUS, with notable hot spots over several of the most populated core-based statistical areas (CBSAs; Fig. 1). On average, RWC was found to contribute 21.4% of the simulated CONUS mean PM_2.5, or 1.08 of 5.05 μg/m³. The maximum grid cell simulated monthly average RWC-related PM_2.5 was 30.14 μg/m³ in Boise, Idaho, with other notable hot spots apparent over Colorado and Minnesota, and the Northeastern US. Population–weighted average RWC-related PM_2.5 concentrations were 2.43 μg/m³, or 21.9% of the overall average simulated population-weighted PM_2.5. Though total PM_2.5 is most often considered for health impact assessments, total PM_2.5 is composed of many individual species. Subspecies analysis of simulated RWC PM_2.5 (table S13) found that the most prevalent RWC PM_2.5 constituent is organic carbon (OC). OC from RWC contributes 0.5 μg/m³ of PM_2.5 on average for CONUS—nearly half (46.2%) of total RWC PM_2.5 concentrations, and a third (33.3%) of total simulated OC concentrations from all sectors.

Fig. 1. Modeled wintertime residential wood combustion (RWC) contributions to fine particulate matter (PM_2.5) across the contiguous United States and major metropolitan areas.

WRF-CMAQ modeled wintertime residential wood combustion (RWC)-related fine particulate matter (PM_2.5) concentrations (μg/m³) for (A) the contiguous United States (CONUS) and (B to J) 9 of the 20 most populated combined statistical areas (CBSA): (B) Seattle-Tacoma-Bellevue, WA; (C) Minneapolis-St. Paul-Bloomington, MN-WI; (D) Chicago-Naperville-Elgin, IL-IN-WI; (E) Boston-Cambridge-Newton, MA-NH; (F) New York-Newark-Jersey City, NY-NJ-PA; (G) Philadelphia-Camden-Wilmington, PA-NJ-DE-MD; (H) Washington-Arlington-Alexandria, DC-VA-MD-WV; (I) Los Angeles-Long Beach-Anaheim, CA; (J) Denver-Aurora-Lakewood, CO. Note that each panel uses an independent color bar. For each CBSA, mean RWC PM_2.5 concentration (μ) and percent of mean PM_2.5 concentration from all sources, population-weighted mean RWC PM_2.5 concentration (μ_p) and percent of population-weighted mean PM_2.5 concentration from all sources, RWC PM_2.5 concentration interquartile range (IQR), and maximum RWC PM_2.5 concentration (max) are provided. The black dashed line in (A) is the 100°W meridian.

On a regional basis, visual inspection of Fig. 1A reveals distinct east-west CONUS RWC PM_2.5 regimes. To investigate each regime, we subdivide our domain along the 100°W meridian, following the guidance of Seager et al. (39). We find that the western CONUS has less overall ambient RWC PM_2.5 but more intense hot spots, especially in densely populated areas. Ambient RWC PM_2.5 concentrations were higher in the eastern CONUS, where at least 1 μg/m³ is simulated in ~75% of grid cells, while just ~9% of grid cells in the western CONUS surpass this threshold. Though the eastern CONUS has ~3× the average RWC PM_2.5 concentrations as the western CONUS (East: 1.51 μg/m³, West: 0.56 μg/m³), the western CONUS has higher population-weighted RWC PM_2.5 concentrations (East: 2.38 μg/m³, West: 2.62 μg/m³). Population-weighted concentrations reflect total exposures; thus, while area-averaged RWC PM_2.5 concentrations in the eastern CONUS are higher, RWC PM_2.5 in the western CONUS may be more impactful.

East-west differences are largely the result of population and emissions density. Average RWC PM_2.5 emissions in the East are higher than in the West: eastern CONUS grid cells emit ~0.21 tons of total January primary RWC PM_2.5 emissions on average, whereas western CONUS grid cells only emit ~0.06 tons on average. However, we find that urban grid cells, defined as those with population densities exceeding 500 residents per square mile, emit an average of 1.15 tons in the western CONUS, compared to 0.90 tons in the eastern CONUS. In addition, other factors, such as topography, surface roughness, ventilation, and accumulation (40–42), have been shown to affect pollutant transport, potentially reducing PM_2.5 dispersion and contributing to PM_2.5 concentrations in urban areas in the western CONUS, which tend to have more complex terrain compared to the East. Topographic influences on simulated PM_2.5 are visible throughout CONUS (Fig. 1A) but are particularly pronounced in California’s Central Valley, across southern Idaho’s Snake River Plain, and in Washington State, where RWC PM_2.5 emissions from communities adjacent to Puget Sound are not transported over the Cascade Mountain Range.

Despite the perception that wood burning for home heating is a rural phenomenon, our simulations find that RWC PM_2.5 hot spots are most prevalent in urban areas, especially in the Western CONUS. Urban areas experienced average RWC PM_2.5 concentrations of 2.47 μg/m³ (22.7%), while rural cells had mean concentrations of just 1.04 μg/m³ (18.9%). In addition to our CONUS-scale urban/rural findings, we note major differences between RWC PM_2.5 concentrations in various major urban areas, i.e., CBSAs. In Fig. 1 (B to J), we plot average winter RWC PM_2.5 concentrations in nine major CBSAs across different geographical regions of the CONUS. The Philadelphia CBSA was simulated to have the highest CBSA-average PM_2.5 from RWC [3.90 μg/m³ (30.4%), PW: 4.52 μg/m³ (29.2%)] (Table 1), while the highest CBSA-average population-weighted RWC PM_2.5 concentrations were simulated in the Denver CBSA [1.64 μg/m³ (34.8%), PW: 6.22 μg/m³ (39.5%)]. Notably, area-average and population-weighted average RWC PM_2.5 concentrations can vary substantially for CBSAs. For example, the Denver CBSA includes large areas with low population density and low PM_2.5 concentrations, resulting in a substantial difference in these metrics (1.6 versus 6.2 μg/m³; Fig. 1J). Each CBSA subplot in Fig. 1 contains grid cells with maximum RWC PM_2.5 concentrations exceeding 4 μg/m³, highlighting the prevalence and magnitude of pollution from RWC in major urban areas. Population-weighted mean concentrations were higher than mean concentrations in each CBSA, indicating that within major CBSAs, RWC PM_2.5 concentrations are more often located in densely populated areas, further contradicting the notion that RWC pollution is a mostly rural phenomenon.

Table 1. Estimated winter RWC PM_2.5 concentrations, population-weighted concentrations, and the attributable mortality for major core-based statistical areas.

See table S14 for further statistics.

CBSA	RWC PM2.5 and percent total PM2.5	Population-weighted mean RWC PM2.5 and percent of total population-weighted mean PM2.5	Attributable mortality rate (per 100k) with 95% CI
Philadelphia	3.90 (30%)	4.52 (29%)	7.6 (5.8, 8.5)
Denver	1.64 (35%)	6.22 (40%)	7.6 (5.7, 8.5)
Minneapolis	3.77 (33%)	5.47 (35%)	7.3 (5.5, 8.1)
Seattle	1.99 (42%)	4.84 (44%)	6.8 (5.2, 7.6)
Boston	3.69 (41%)	3.88 (35%)	5.0 (3.8, 5.6)
Washington, DC	3.43 (31%)	4.12 (29%)	4.7 (3.6, 5.3)
New York City	2.94 (28%)	3.15 (21%)	4.2 (3.2, 4.7)
CONUS	1.08 (21%)	2.43 (22%)	4.0 (3.0, 4.5)
Los Angeles	1.44 (17%)	2.67 (16%)	3.4 (2.6, 3.8)
Chicago	1.86 (18%)	2.12 (16%)	3.3 (2.5, 3.7)
Phoenix	0.98 (17%)	1.62 (16%)	2.7 (2.1, 3.1)
Atlanta	1.65 (14%)	1.80 (14%)	2.6 (2.0, 3.0)
Houston	1.08 (10%)	1.30 (9%)	1.8 (1.3, 2.0)
Dallas	0.99 (10%)	1.10 (9%)	1.5 (1.1, 1.7)
Miami	0.35 (6%)	0.39 (5%)	0.6 (0.5, 0.7)

Adding to the complexity of our CBSA-level findings, we find substantial heterogeneity of simulated RWC PM_2.5 concentrations within CBSAs. Average RWC PM_2.5 concentrations in eastern and midwestern CBSAs (Fig. 1, C to H) vary spatially but have weaker gradients than western CBSAs. In contrast, CBSAs such as Seattle, Los Angeles, and Denver exhibit enhanced spatial variability in simulated concentrations (Fig. 1, B, I, and J). These were the only three CBSAs modeled to have interquartile concentration ranges (IQRs) greater than the mean concentration of the domain and have visually apparent steep RWC PM_2.5 concentration gradients. Each of the three CBSAs contains mountain ranges within their respective boundaries: the Rocky Mountains west of Denver, the Cascade Range to the east of Seattle, and the San Gabriel Mountains north of Los Angeles. The topography in these areas limits pollutant dispersion, contributing to the heterogeneous simulated concentrations within the CBSA, and potentially increasing long-term average RWC PM_2.5 concentrations in these cities (43–46).

The spatial patterns within CBSAs reflect both local emissions and atmospheric processes. In addition to the primary PM_2.5 emitted during RWC, secondary PM_2.5 forms and evolves through chemical reactions in the atmosphere from gaseous precursors such as volatile organic compounds and nitrogen oxides (47). Primary PM_2.5 and precursors of secondary PM_2.5 are also transported and dispersed by meteorological and topographic conditions, shaping the spatial variability of RWC PM_2.5 concentrations observed across CBSAs.

We note that even CBSAs in what are traditionally thought of as “warm climates” are simulated to have RWC impacts. For example, in our simulations, the Los Angeles (LA) CBSA has substantial RWC PM_2.5 emissions, which are distributed across the CBSA’s sprawling residential areas (fig. S17). Despite its warm-climate reputation, our simulated LA CBSA temperatures fall below the heating degree day (HDD) threshold of 10°C (50°F) (figs. S19 and S20) for parts of our simulation period, which leads to modeled RWC activity and associated emissions. We find substantial simulated average RWC PM_2.5 concentrations in the LA CBSA [1.43 μg/m³ (16.8%), PW: 2.67 μg/m³ (16.3%)], including more than 3 μg/m³ in downtown LA, with a hot spot of 4.7 μg/m³ in the residential Pasadena neighborhood north of the LA city center. Although LA does not exhibit RWC-related PM_2.5 concentrations as high as those observed in cities with colder climates, the average RWC exposure of 2.67 μg/m³ remains substantial, suggesting that RWC impacts may be prevalent even in cities with temperate climates.

Differences in emissions and concentrations—An NYC CBSA example

Processed PM_2.5 emissions and simulated concentrations were found to be only moderately spatially correlated (r = 0.58) in the New York City (NYC) CBSA, suggesting that PM_2.5 emissions may not be a good predictor of exposures. Though RWC PM_2.5 emissions and concentrations are generally correlated, spatial distributions and correlation strengths vary by CBSA (fig. S22). Here, we use the NYC area to illustrate these differences. Emissions (Fig. 2A) are concentrated in the Yonkers and Long Island suburbs, north and east of the city, and little to no emissions are modeled in the densest boroughs. However, simulated RWC PM_2.5 concentrations in the NYC CBSA (Fig. 2B) are more uniformly dispersed across the CBSA domain than emissions, contributing ~28% of total PM_2.5 and ~21% of estimated winter exposures [2.69 μg/m³ (28.3%), PW: 3.15 μg/m³ (21.1%)]. Furthermore, RWC PM_2.5 concentrations exceed 2 μg/m³ in ~77% of the CBSA, including the densely populated city center that contributes only limited RWC PM_2.5 emissions. Though the largest concentrations of PM_2.5 from RWC are modeled on the outskirts of the NYC CBSA, primary emissions from hot spots well outside of the city center lead to elevated concentrations in the most densely populated regions, resulting in a poor overall correlation (Fig. 2C). This finding highlights the utility of using air quality models with dynamically coupled meteorology and atmospheric chemistry when analyzing pollution impacts: spatial distributions of emissions and concentrations can differ substantially (table S6).

Fig. 2. Differences in New York City (NYC) combined statistical area (CBSA) RWC emissions and concentrations.

Plotted are the NYC combined CBSA RWC (A) total January estimated primary fine particulate matter (PM_2.5) emissions in metric tons, (B) average winter WRF-CMAQ simulated PM_2.5 concentrations, and (C) scatter plot of estimated primary PM_2.5 emissions and winter-average WRF-CMAQ simulated PM_2.5 concentrations. The red line is the best-fit linear regression through the origin (0,0), which provides a simple visual relationship between emissions and concentrations despite different units. See fig. S21 for plots of other CBSAs.

RWC PM_2.5-attributable health impacts

To assess the health impacts associated with wintertime RWC PM_2.5 exposure, we estimate RWC PM_2.5-attributable premature mortalities at the census tract level. We estimate ~8600 (95% CI: 6500 to 9600) premature deaths per year are associated with winter RWC PM_2.5, constituting ~22% of the total premature deaths attributable to winter PM_2.5 concentrations in our baseline simulation [~38,900 (95% CI: 29,500 to 43,500)].

At the CBSA level, Philadelphia and Denver have the highest attributable mortality rates, while Denver has the highest population-weighted concentration (Table 1). Five southern cities—Miami, Phoenix, Atlanta, Dallas, and Houston—cities not typically associated with RWC—are estimated to have the lowest attributable mortality rates. However, PM_2.5 from RWC was still estimated to cause a combined ~360 (95% CI: 270 to 400) annual premature deaths in these CBSAs.

Distributional impacts

We analyze CONUS-level population average RWC PM_2.5 concentration exposure, attributable mortality rates, and baseline mortality rates (BMRs) and find that relative disparities in RWC health impacts are driven by both differences in underlying BMRs and RWC PM_2.5 concentrations (Fig. 3A). The average census tract in CONUS has an RWC PM_2.5 concentration of 2.51 μg/m³, a BMR of 1.08%, and 4.23 deaths per 100,000 attributable to wintertime RWC PM_2.5. Because total attributable mortality depends on both pollution exposure and the baseline vulnerability of the population (as captured by BMRs), disparities in RWC PM_2.5 health impacts can differ from disparities in exposure. For example, despite being exposed to CONUS-average RWC PM_2.5 concentrations, we find that the non-Hispanic white population in the CONUS experiences ~4% lower RWC-attributable mortality rates relative to the CONUS average, driven by ~3% lower than average BMRs. However, the Black population, also exposed to approximately average RWC PM_2.5 concentrations (~1% lower), experiences mortality rates attributable to RWC PM_2.5 that are ~8% higher than the CONUS average. This is primarily due to BMRs that are ~7% higher than the CONUS average. In contrast, the Asian population was exposed to ~17% higher than average RWC PM_2.5 concentrations; however, we estimate ~12% lower RWC-attributable mortality rates relative to CONUS averages due to ~28% lower than average BMRs. Though the Native Hawaiian or Pacific Islander population has ~14% lower than average BMRs, we estimate ~22% higher RWC PM_2.5 concentrations, resulting in ~7% elevated RWC-attributable mortality rates.

Fig. 3. Racial/Ethnic distributional impacts and relative disparities.

(A) A comparison of RWC fine particulate matter (PM_2.5) concentrations (shaded), baseline mortality rates (BMRs) (boxed), and estimated RWC PM_2.5-attributable mortality rates (dot) for each respective racial/ethnic group relative to the CONUS averages (in %). Positive values indicate higher-than-average concentrations or attributable mortality rates, while negative values indicate lower-than-average concentrations or attributable mortality rates. (B to I) Population distribution of race/ethnicity that comprises each decile of RWC PM_2.5 concentrations (μg/m³, upper) and associated attributable mortality rates (annual deaths per 100,000, lower) for (B) CONUS and three combined statistical areas (CBSAs): (C) Los Angeles-Long Beach-Anaheim, CA; (D) Chicago-Naperville-Elgin, IL-IN-WI; and (E) New York-Newark-Jersey City, NY-NJ-PA. The y axis shows the range of RWC PM_2.5 concentrations within each decile, and the x axis represents population percentages. All racial/ethnic groups, except Hispanic, include only those identifying as non-Hispanic. See fig. S22 for further CBSA-level relative disparity plots and fig. S23 for further CBSA-level decile plots.

We also examine distributional differences and disparities at a more granular level. That is, in Fig. 3 (B to I), we plot the distribution of population fractions, i.e., the proportion of a racial/ethnic population demographic within a census tract, for each decile of RWC PM_2.5 concentrations and their associated attributable mortality rates for the CONUS and select CBSAs. To quantify trends and support the robustness of our findings, we also compute correlations between nonwhite population fractions and RWC PM_2.5 concentrations and attributable mortality rates (table S15). A positive correlation coefficient implies that census tracts containing larger populations of people of color are associated with increased RWC PM_2.5 concentrations or mortalities, and the magnitude of the coefficient describes the strength of the trend.

At the CONUS level, the racial/ethnic distribution of concentrations of RWC PM_2.5 and associated mortality rates largely reflects the overall CONUS-level racial/ethnic population distribution, shown at the bottom of Fig. 3 (B to I). However, as mortality rate deciles increase, the Black population fraction also increases (dark blue sections, Fig. 3F). This increase reflects elevated BMRs (Fig. 3A), which yield higher attributable mortality for a given exposure (Eq. 2), consistent with the findings of Josey et al. (48).

Within each CBSA subplot (Fig. 3, C to E and G to I), white population fractions (light blue sections) decrease at higher RWC-related PM_2.5 concentrations and attributable mortality rates, suggesting people of color face increased pollution burdens from RWC. The Los Angeles CBSA exhibits particularly strong patterns of increasing nonwhite population fractions at both elevated RWC PM_2.5 concentrations (Fig. 3C, r = 0.26) and attributable mortality rates (Fig. 3G, r = 0.25), as shown by the decreasing white population fraction at higher deciles. Similarly, in the Chicago CBSA, increasing nonwhite population fractions are seen at both higher RWC PM_2.5 concentrations (Fig. 3D, r = 0.23) and higher attributable all-cause mortality rate deciles (Fig. 3H, r = 0.41). In the Chicago CBSA, the Black population fraction shows a strong association with RWC-attributable mortality rates (blue bars, Fig. 3H, r = 0.55) while having low correlation with RWC PM_2.5 concentration deciles (blue bars, Fig. 3D, r = 0.05), exemplifying the impact of BMRs on health impacts. Similarly, the NYC and Los Angeles CBSAs do not indicate increasing nonwhite populations at higher RWC PM_2.5 concentrations (r = 0.05 and r = 0.04, Fig. 3E), yet deciles of increasing mortality rates show increasing nonwhite population fractions (r = 0.17 and r = 0.18, Fig. 3I). The elevated RWC PM_2.5 attributable mortality rates for Black populations in Chicago, NYC, and Los Angeles in the absence of higher RWC PM_2.5 exposure highlights the substantial role of underlying susceptibilities in driving public health outcomes.

Given the substantial role of underlying susceptibilities, we perform an additional sensitivity analysis, in which we use the race/ethnicity-specific concentration-response function from Di et al. (49), adapting the methods of Geldsetzer et al. (50). We find that total estimated RWC-attributable mortality increased ~55% relative to our Chen and Hoek (51)–based analysis (table S9). In addition, use of the Di et al. (49) race/ethnicity-specific concentration-response function strengthened simulated disparities, with substantially higher mortality burdens in census tracts with larger nonwhite populations due to the higher relative risks for people of color (fig. S27).

DISCUSSION

This study aimed to quantify the air pollution and attributable mortality associated with RWC-related PM_2.5 exposure and examined the associated relative disparities in exposure and mortality across different population groups. This study was motivated by the 2023 update to the RWC emissions by the NEI, as well as a new 4-km CONUS-wide two-way coupled WRF-CMAQ simulation. We found that more than 20% of both population-weighted and unweighted winter PM_2.5 concentrations are attributable to residential wood burning, and we estimate that winter RWC PM_2.5 causes ~8600 premature deaths annually. In addition, we found that the nonwhite population fraction is positively correlated with RWC PM_2.5 concentrations in 18 of the CONUS’ 20 most populous CBSAs and positively correlated with RWC-attributable PM_2.5 mortality in 17 (table S15). Notably, however, in 15 of these CBSAs, the nonwhite population fraction is correlated with lower RWC emissions (table S16). These findings indicate that in the largest metropolitan areas of the CONUS, people of color are disproportionately burdened by RWC PM_2.5, despite contributing below-average emissions. We found that the magnitude of this result was sensitive to our chosen concentration-response function (table S18). Our sensitivity analyses using race/ethnicity-specific concentration-response functions from Di et al. (49) amplified the relative disparities identified in our Chen and Hoek–based findings (see discussion in the Supplementary Materials; fig. S27 and tables S18 and S19). We note however, that the application of the Di et al. (49) methodology required an assumption that their concentration-response function, which was derived from a 65 and older cohort, could be used to estimate mortalities of all adults who are more than 25 years old, as assumed by Geldsetzer et al. (50).

Our estimate of ~8.6k winter RWC PM_2.5 premature deaths is likely higher than the estimates reported by Penn et al. (37), i.e., ~10k annually across all seasons. Several factors contribute to differences in mortality estimates: (i) Penn et al. (37) used 2005 NEI RWC emissions estimates of ~370,000 tons per year, whereas here, we use the 2020 updated NEI emissions estimate of ~485,000 tons per year. (ii) Penn et al. (37) also used an air quality model that resolved pollutant concentrations at a spatial resolution of 36 km, while our simulation has a resolution of 4 km. (iii) Penn et al. (37) used county-level baseline mortality data, whereas we use finer-resolution census tract-level data. Higher-resolution models and health calculations often result in higher attributable mortality because higher concentrations of RWC PM_2.5 are generally found in more populated regions and areas with more susceptible populations. For example, Li et al. (52) found that air quality models run at finer resolutions estimate higher attributable mortality from PM_2.5 in the US, especially for primary PM_2.5 species. Likewise, Korhonen et al. (53) found that simulating primary PM_2.5 at a 50-km resolution produced 55% lower mortality than simulations at 5 km, in a Finland-focused study. In addition, epidemiological studies have identified high-fidelity, high-resolution air quality models to be better at capturing pollution-related health impacts than coarser alternatives (54–56).

Validation of RWC PM_2.5 findings is difficult because typical ground monitoring station data, such as the EPA AQS used to validate our model, measure either total or speciated PM_2.5. RWC is one of many sources of each species of PM_2.5, so the RWC sector contribution cannot immediately be discerned from the data. To further assess the accuracy of our simulated RWC PM_2.5 concentrations, we compared findings to recent source apportionment studies across the US (table S12). RWC PM_2.5 concentrations and fractions of total PM_2.5 were generally well aligned with available source apportionment sites throughout the Pacific Northwest, Southeast US, and in major cities such as NYC and Los Angeles. However, simulated grid cell RWC PM_2.5 concentrations severely underestimated source apportionment measurements in two small, rural towns (Lakeview, OR and Oakridge, OR). Estimated total January RWC PM_2.5 emissions in these grid cells were 0.004 and 0.044 metric tons, respectively. Because these towns each have populations under 4000, the unit housing density-based surrogates used to spatially allocate emissions assigned little or no RWC emissions to these grid cells. As a result, WRF-CMAQ simulated negligible RWC PM_2.5 concentrations in these locations.

Our analysis of RWC emissions and adverse health impacts is subject to several limitations and areas for future improvement. First, the emissions data for RWC remain estimates, which inherently introduce uncertainties into the results. The use of 2020 RWC emission inventories applied within the 2016 beta emission modeling platform resulted in emission species that were omitted due to the misalignment in aerosol modules, potentially resulting in an underestimation of modeled concentrations (table S3). Second, owing to the computational intensity of 4-km CTM modeling for the CONUS domain, our study uses a single month of RWC results, which underestimates the variability of winter RWC activity and the influence of meteorological variability on where and when pollutants accumulate. Furthermore, the minimum simulation layer height of ∼20 m is likely an overestimation of typical RWC emission heights, introducing additional uncertainty. We also quantified the health impacts of RWC during the winter season only, which is an underestimate of the total health impacts caused by RWC activity during all seasons. Last, we do not account for potential increases in indoor exposure among individuals near sources of RWC. With additional computer power and resources, future research could expand the temporal scope of the study to include more representative months, which would provide a more robust estimate of annual health impacts, rather than only wintertime health impacts. Including a summer month could better model RWC emissions from other seasonal sources, such as outdoor recreational burning, adding greater nuance to the analysis.

Our use of a 4-km resolution for health impact assessments represents considerable progress, enabled by recent advancements in computational modeling. We aggregated simulated PM_2.5 to the commonly used 12- and 36-km resolutions (figs. S24 to S26) and found that the coarser resolutions decreased RWC-attributable mortality estimates by ~4% for 12 km and ~14% for 36 km, consistent with recent published findings (57). Underestimations with aggregated 12- and 36-km RWC PM_2.5 simulation data were more pronounced at the CBSA level, decreasing 4-km estimated attributable premature mortalities by 40% in some areas (table S17). Pollutant concentrations resolved at finer scales, such as 1 km, or even 250 m, have been shown to substantially affect population exposure estimates and health impact assessments, especially in dense urban areas (53, 58). However, while higher-resolution simulations are theoretically ideal, key constraints include the lack of high-quality, highly resolved input data to initialize and constrain the model, the challenge of sourcing computational resources to run high spatiotemporal simulations, and the lack of observational data to validate the performance of high-resolution models.

Despite the above caveats, our study presents a state-of-the-science update of the impacts of RWC in the CONUS. Future research needs include the further refinement of emissions inventories to best represent the location and magnitude of emissions. Larger-scale survey studies would enable the continued improvement of inventory accuracy of a variety of sectors, including RWC, and exploring advanced modeling frameworks could further reduce uncertainties and provide a more complete assessment of emissions (59). In addition, we found only limited meta-analyses of the concentration-response effects of PM_2.5 derived from RWC. Note that RWC refers specifically to household wood combustion for heating, whereas biomass burning also includes wildfires and agricultural burning, which differ in emissions characteristics. Karanasiou et al. (28) conducted a meta-analysis on the short-term health effects of exposures to PM_2.5 and PM₁₀ from biomass burning, Chen et al. (60) conducted a meta-analysis on long-term health effects of exposures to common constituents of PM_2.5, Ma et al. (61) and Qiu et al. (62) derived long-term nonlinear exposure response functions to PM_2.5 from wildfire smoke, and various meta-analyses studied specific health outcomes from indoor exposure to solid fuel combustion (63–65). However, to our knowledge, no systematic review and meta-analysis of the long-term mortality impacts of RWC PM_2.5 exists, leading studies estimating long-term health impacts of RWC PM_2.5 (66, 67) to use generic PM_2.5 concentration-response functions. Further research on the long-term health impacts of exposure to PM_2.5 from wood-burning sources could improve the accuracy of RWC health impact estimates by providing woodsmoke-specific toxicity and associated relative risk values.

Our findings underscore the substantial air quality and health impacts of RWC and the considerable potential for pollution reductions through targeted mitigation efforts, particularly in urban and suburban areas. Wintertime RWC accounted for sizable proportions of population-weighted PM_2.5 concentrations across all major cities analyzed (Table 1), including those with warmer climates and limited wood-burning activities. Notably, RWC emissions often originate outside city centers, yet the most densely populated areas often experience the highest attributable mortality rates due to pollution transport to populations with higher baseline susceptibilities. Our results suggest that addressing the health burdens of RWC exposure requires policy interventions at the CBSA or larger scale, as RWC-related PM_2.5 is transported across state boundaries, especially in multistate metro areas, such as NYC, Philadelphia, and Washington, DC. Our findings indicate that targeted RWC emission mitigation strategies aimed at reducing RWC in regions with poor wintertime air quality could be a resource-efficient approach to improve air quality, reduce premature deaths from air pollution, and mitigate unequal pollution burdens.

MATERIALS AND METHODS

Experimental design

To estimate the CONUS-level contribution of RWC to PM_2.5 concentrations, we compare a baseline WRF-CMAQ simulation to a sensitivity simulation, no-RWC, in which all RWC emissions are eliminated. To perform WRF-CMAQ simulations, we follow the methodology of Wong et al. (68): (i) we produce dynamically downscaled meteorology with stand-alone WRF simulations, (ii) we then use the stand-alone WRF outputs to create meteorologically informed emissions data using the Sparse Matrix Operating Kernel of Emissions (SMOKE), and, lastly, (iii) we run the coupled WRF-CMAQ model, incorporating the meteorologically informed SMOKE emissions data.

Air quality model setup and RWC emissions estimates

To generate boundary and initial conditions and facilitate the production of meteorologically informed emissions data, we first perform stand-alone WRF simulations to generate three-dimensional meteorology with a 4-km resolution CONUS domain nested within a 12-km CONUS domain (fig. S2). We use a 10-day spin-up period and simulate the month of January 2016 using 24-s time steps for the 4-km domain. We focus our analysis of RWC impacts on the winter season because RWC activity is highest during colder months (69). Given the more than 200,000 core hours required for WRF-CMAQ to simulate dynamic coupled meteorology and atmospheric chemistry at 4 km over the CONUS, here we use January as a proxy for the winter season. While this assumption is limiting, it is common in recent literature (70, 71) to use a single month to represent the winter season in residential biomass burning impact analyses. In addition, we find that January 2016 had similar PM_2.5 spatiotemporal concentrations to other 2010 to 2024 Northern Hemisphere winter months (figs. S3 and S4).

Stand-alone meteorological modeling

We use the soil moisture initialization option during the 10-day spin-up to allow soil moisture and soil temperature variables to reach a state of statistical equilibrium with observational constraints (72). We run WRF with 35 vertical layers from the surface to 30 hPa with a lowest model level thickness of ∼20 m. Initial conditions and 3-hour lateral boundary conditions for the 4-km domain are sourced from the North Atlantic Regional Reanalysis (73) and chemical boundary and initial conditions from CAM-Chem (74, 75). Simulated WRF meteorology is nudged toward reanalysis using Four-Dimensional Data Assimilation above the boundary layer, using nudging coefficients from (76) and (77) for temperature and wind (1 × 10⁻⁵ for 4 km) and the water vapor mixing ratio (1 × 10⁻⁶ for 4 km). We incorporate the land cover product from the National Land Cover Database (78) at a 9–arc sec resolution. For the WRF physics options, we select the Morrison 2-moment microphysics scheme (79), version 2 of the Kain-Fritsch (KF2) cumulus cloud parameterization for the 12- and 4-km simulations (80), the Asymmetric Convective Model version 2 for the planetary boundary layer (81), and the Pleim-Xiu land surface model (72) with soil moisture and temperature nudging (82, 83). We use the Rapid Radiative Transfer Model for GCMs (84) for both our shortwave and longwave radiation schemes.

Emissions modeling

Meteorologically informed emissions were generated for the 4-km baseline and no-RWC scenarios over the CONUS using the SMOKE, v4.5 processing system (85) with the EPA 2016 Beta platform (86). We use spatial and temporal surrogates to process RWC county-level annualized emissions to the subcounty fixed 4-km grid cells needed to run WRF-CMAQ. The spatial surrogates distribute RWC emissions based on the density of detached homes (used for pellet-fired stoves, pellet-fired furnaces, pellet-fired hydronics, outdoor hydronics, and outdoor firepits) and the density of single and dual-unit homes (used for all other wood-burning devices; fig. S5) (87). These surrogates were developed using 5-year 2016 to 2020 survey data from the American Community Survey (ACS) (12). Emissions were distributed temporally using a regression algorithm developed and validated to relate RWC tracers to outdoor temperature (88). The algorithm only allocates emissions to days with minimum temperatures below 10°C (50°F) to capture seasonal fluxes in RWC activity (89, 90).We improved upon the representation of RWC in the Beta platform by upgrading the RWC-specific emissions input, ancillary data, speciation scripts, and surrogates using components from the NEI 2020 emissions modeling platform (88). The 2020 NEI calculates RWC emissions using the following equation

$$E_{\mathrm{c},\text{SCC},\mathrm{p}}=H_{\mathrm{c}}\times {\text{AF}}_{\mathrm{c},\mathrm{a}}\times {\text{BR}}_{\mathrm{c},\mathrm{a}}\times D_{\mathrm{c}}\times {\text{DP}}_{\text{SCC}}\times {\text{SAF}}_{\mathrm{s}}\times {\text{HAF}}_{\mathrm{c}}\times {\text{EF}}_{\text{SCC},\mathrm{P}}$$ (1)

where E_{c, SCC, p} is the annual total emissions for a given RWC appliance subclassification, county, and pollutant; H_c is the number of occupied homes in each county; AF_{c, a} is the fraction of homes in each county using each RWC appliance; BR_{c, a} is the average annual amount of wood burned per appliance for each county; D_c is the density of firewood per county; DP_SCC is the distribution profile of the RWC appliance to source classification codes (SCCs), which contain information such as EPA certification, age, and catalytic versus noncatalytic; SAF_s is the state total adjustment factor derived from State Energy Data System (SEDS) annual emissions estimates; HAF_c is the housing density adjustment factor per county; and EF_{SCC, P} is the emissions factor in quantity of pollutant emitted per ton of wood burned.

The 2020 NEI models RWC based on county-level wood-burning activity data from a 2018 national survey conducted by the US EPA, Commission on Environmental Cooperation (figs. S6 and S7), the Northeast States for Coordinated Air Use Management, and Abt Associates (10). Appliance fractions and burn rates were created for each county using logistic regressions considering home type, urban/rural, number of HDDs, population density, percent forest cover, and percent of homes using natural gas as the primary heating fuel in each county. Statewide emissions totals were normalized to the 2020 Energy Information Administration’s SEDS data (91). The 2020 NEI represents a considerable update to the methodologies used to develop RWC emissions inventories in the NEI 2005 and NEI 2014 versions, which used coarser census region and division input data and omitted forest cover from regressions. See table S2 for details on the 2020 and 2014 NEI methodologies. In addition to the new methods used to simulate the RWC emissions inventory, the 2020 inventory used an updated aerosol mechanism AE7 with ancillary data and scripts designed for the SMOKE, v5.5 environment, which required remapping to be compatible with the Beta platform and AE6 Beta framework (table S3).

Coupled air quality and meteorology modeling

Having generated meteorologically informed, processed emissions (fig. S8 and tables S4 to S6), we then simulate PM_2.5 concentrations using the two-way coupled CMAQ (v5.2) (92) and WRF (v3.8) (93) modeling system (WRF-CMAQ) (68), allowing feedback between the WRF radiation scheme and CMAQ simulated aerosols. The CMAQ model uses emissions and meteorology data to model atmospheric chemistry, physics, transport dynamics, and secondary interactions to predict atmospheric pollutant concentrations. CMAQ is configured with the Carbon Bond version 6 (CB6) gas-phase mechanism and aerosol module version 6 (AE6) aerosol mechanism with aqueous chemistry (CB6r3_AE6_AQ). CMAQ simulates PM_2.5 by individual chemical species, which are then summed to obtain the total mass concentration of PM_2.5. To determine the contribution of RWC to PM_2.5 concentrations and assess the downstream population exposure, health outcomes, and distributional impacts, we subtract the no-RWC simulation from the baseline simulation.

Model validation and performance metrics

To assess the performance of two-way coupled WRF-CMAQ simulated meteorology (fig. S9), we compared simulated meteorological variables to ground-based measurements (table S7). We use hourly ground-based Local Climatological Data (LCD) from the National Climatic Data Center (NCDC) as the validation dataset. Following the recommendations of Emery and Tai (94), we evaluate 2-m temperature, relative humidity at 2 m, and wind speed and wind direction at 10 m because each variable is relevant to the transport and chemical modeling of atmospheric air pollutants. Meteorological variables are evaluated by comparing the simulated meteorology at each 4-km grid cell containing an NCDC station to the station measurements. The CONUS includes 224 stations, permitting model evaluation at 0.04% of the ~600,000 land-covering grid cells (figs. S10 to S13). Wind direction comparisons exclude readings at wind speeds under 0.5 m/s, which are reported with no wind direction. Model fidelity statistics align with performance recommendations and are similar to previously published, though coarser-resolution WRF-CMAQ CONUS studies (table S8).

We compared simulated PM_2.5 concentrations (fig. S14) to the EPA Air Quality Station (AQS) hourly surface observations using metrics suggested by Dennis et al. (95) (table S9) to assess the accuracy of WRF-CMAQ PM_2.5. Normalized metrics are used to allow for intercomparisons between studies covering different domains and levels of pollution (table S11). Modeled concentrations are compared to surface observations in each of the 370 grid cells containing an AQS station, 0.06% of the total land-covering grid cells (fig. S15). Simulated RWC-related PM_2.5 concentrations were also compared to source apportionment studies to validate the RWC portion alone (table S12).

Health impacts and relative disparities

To estimate the health impacts attributable to RWC-related PM_2.5 concentrations in winter, we estimate the all-cause mortality associated with RWC PM_2.5 exposure at the census tract level. We define RWC-related PM_2.5 concentrations to be the difference between simulated PM_2.5 concentrations in the baseline and no-RWC scenarios. Census tract PM_2.5 concentrations were determined by calculating the area-weighted average WRF-CMAQ simulated PM_2.5 concentrations of the 4-km grid cell intersections within each census tract using geographic shapefiles (IPUMS 2024). The epidemiologically derived relative risk (RR) for long-term exposure is used to estimate the premature all-cause mortality attributable to wintertime average RWC PM_2.5 concentrations (Eq. 2)

$${\text{Mort}}_{\text{RWC},\text{CT}}={\text{MR}}_{\text{CT}}·{\text{Pop}}_{\text{CT}}·[1-\text{exp}(-\mathrm{\beta }{\mathrm{\Delta }x}_{\text{CT}}) ]$$ (2)

where Mort_{RWC, CT} is the total all-cause mortality attributable to RWC PM_2.5 concentrations for each census tract, MR_CT is the age-stratified census tract-level baseline all-cause mortality rate, Pop_CT is the age-stratified census tract population, β is the concentration-response coefficient (51), and ${\mathrm{\Delta }x}_{\text{CT}}$ is the census tract annual-average RWC-contributed PM_2.5 concentration, estimated to be wintertime (December to February) RWC-related PM_2.5 concentrations (baseline − no-RWC) divided by 4 because we neglect RWC pollution during all other seasons. Census tract-level baseline all-cause age-stratified mortality rates were obtained from Industrial Economics Inc., derived from USALEEP abridged life tables modified for national health impact analyses (96). Census tract age-stratified population data were retrieved from the 5-year averaged (2015 to 2019) ACS data (97). Four-kilometer gridded population totals used for population-weighted concentration calculations were obtained from the Beta platform. We use a widely used PM_2.5 all-cause mortality RR of 1.08 (95% CI: 1.06 to 1.09) per 10 μg/m³ increase for adults who are more than 25 years old, derived from a meta-analysis of 25 cohort studies (51). As such, we only analyzed impacts for adults who are more than 25 years old, and we define the RWC PM_2.5 attributable mortality rate as deaths per 100,000 adults who are more than 25 years old. While the RR used is derived from annual PM_2.5 concentrations and is therefore associated with long-term exposures, we apply the same RR to the wintertime concentrations simulated in this study as specific RRs of PM_2.5 associated with seasonal all-cause mortality are limited.

In addition to our analysis based on the RR from (51), we conducted a sensitivity analysis using the race/ethnicity-specific RRs from Di et al. (49), applying ACS census-tract population counts by race/ethnicity to estimate attributable mortality. Though the Di et al. (49) concentration-response function was derived from Medicare users, all of whom are more than age 65, we adapt the methods of Geldsetzer et al. (50) and apply the RRs to all adults who are more than 25 years old. The Di et al. (49) concentration–response function is stepwise linear, with no effect between 0 and 5 μg/m³ and a positive, race-specific slope above 5 μg/m³. Because we do not estimate annual total PM_2.5, we assume annual averages are above this 5 μg/m³ threshold so that RWC-related increments fall within the linear portion of the function. This assumption is supported by the fact that 93% of the CONUS population in this study experiences January-average PM_2.5 levels above 5 μg/m³. Given these assumptions and their inherent uncertainties, we present this sensitivity analysis in the Supplemental Materials. For further details, see the Supplementary Materials (“Health Impact Assessment Sensitivity” section, fig. S27, and tables S18 and S19).

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S27

Tables S1 to S20

References