Wildfires Increase Concentrations of Hazardous Air Pollutants in Downwind Communities

Abstract

Due in part to climate change, wildfire activity is increasing, with the potential for greater public health impact from smoke in downwind communities. Studies examining the health effects of wildfire smoke have focused primarily on fine particulate matter (PM_2.5), but there is a need to better characterize other constituents, such as hazardous air pollutants (HAPs). HAPs are chemicals known or suspected to cause cancer or other serious health effects that are regulated by the United States (US) Environmental Protection Agency. Here, we analyzed concentrations of 21 HAPs in wildfire smoke from 2006 to 2020 at 309 monitors across the western US. Additionally, we examined HAP concentrations measured in a major population center (San Jose, CA) affected by multiple fires from 2017 to 2020. We found that concentrations of select HAPs, namely acetaldehyde, acrolein, chloroform, formaldehyde, manganese, and tetrachloroethylene, were all significantly elevated on smoke-impacted versus nonsmoke days (P < 0.05). The largest median increase on smoke-impacted days was observed for formaldehyde, 1.3 μg/m³ (43%) higher than that on nonsmoke days. Acetaldehyde increased 0.73 μg/m³ (36%), and acrolein increased 0.14 μg/m³ (34%). By better characterizing these chemicals in wildfire smoke, we anticipate that this research will aid efforts to reduce exposures in downwind communities.

1. INTRODUCTION

Wildfire activity has been increasing in the western United States (US) and in other areas, due in part to climate change and past fire suppression.^1,2 In the western US, studies have shown a relative increase in total area burned and high severity fires in recent decades,^3–5 and projections show wildfire activity is expected to increase in the US and elsewhere.^6,7 The increase in wildfire activity coincides with an increasing burden of wildfire smoke on public health.^8,9 Wildfires negatively affect air quality as the smoke emitted is a complex mixture containing numerous air pollutants, including carbon dioxide, volatile organic compounds (VOCs), and fine particulate matter (PM_2.5), the pollutant typically of greatest health concern.^10–12 Currently, wildfires account for an estimated 50% of total PM_2.5 concentrations in some areas of the western US due to both a rise in wildfire activity and a concomitant decline in other sources.^13,14

Building on decades of research on ambient PM_2.5 exposures (i.e., exposures during a typical day, not extreme events), there is extensive epidemiologic evidence relying on different exposure indicators (e.g., total or wildfire-specific PM_2.5 and smoke density) that wildfire smoke is associated with a range of health effects including increases in respiratory- and cardiovascular-related emergency department visits, hospital admissions, and even death.^15–17 Therefore, while wildfire smoke is a complex mixture, the use of PM_2.5 has been shown to be a reliable surrogate for smoke exposure.¹⁵ However, as the acreage burned and smoke emitted from wildfires increases, and some wildfires enter the wildland−urban interface (WUI), burning human-made structures and materials (e.g., buildings, vehicles), additional toxic pollutants may be emitted from the combustion of both natural and nontraditional fuels.^12,18 This potential change in the smoke mixture is important to characterize as current approaches to protect public health rely on reducing PM_2.5 concentrations,¹⁹ and it is unclear if currently recommended interventions would be effective at reducing other increased pollutants.^15,20,21

Among pollutants potentially increasing due to the prevalence of wildfire smoke are those classified by the US Environmental Protection Agency (US EPA) as hazardous air pollutants (HAPs). HAPs are regulated under the Clean Air Act and are defined as “pollutants that are known or suspected to cause cancer or other serious health effects”.²¹ HAPs include organic compounds, such as benzene and formaldehyde, and inorganic chemicals, such as lead and manganese. Wildfire smoke emissions have been extensively studied, and there are emissions factors available for individual HAPs.^11,14,20 However, information regarding ambient concentrations of HAPs is more limited. A recent aircraft field campaign measured gas-phase HAPs in fresh and aged wildfire smoke,¹⁰ and a national-scale health impact study demonstrated potential health implications of exposure to HAPs from wildfire smoke.⁸ Furthermore, there have been monitoring studies of HAPs concentrations at localized sites.^22,23 To the best of our knowledge, however, ground-level concentrations of HAPs in wildfire smoke have not been characterized at a regional scale.

To help fill this knowledge gap, we used 15 years (2006 to 2020) of monitoring data from the US EPA’s Air Quality System (AQS)²⁴ to characterize HAPs on smoke-impacted vs nonsmoke-impacted days across the western US. In all, we analyzed data for 21 HAPs from 309 monitoring stations. These stations are in both urban and rural areas at the ground level and thus measure concentrations of HAPs in communities downwind of fire. We further conducted a detailed case study of HAPs in wildfire smoke at a ground-level monitoring site in San Jose, CA. We focused on this site because it measured concentrations in a major population center; allowed a comparison to HAPs from urban-area, nonfire sources; had extensive monitoring data over the study period; and was impacted by multiple nearby wildfires.

Using monitoring data and data from the case studies, we specifically addressed the following research questions:

1. How much do individual HAPs increase on wildfire smoke days compared with nonsmoke days?

2. How do concentrations of HAPs during wildfire smoke events compare with available reference concentrations (RfCs) for acute and chronic human health effects for those chemicals?

By quantifying these HAPs, we hope to improve our understanding of the risk that wildfire smoke poses to public health and aid in efforts to reduce exposure to these chemicals.

2. DATA AND METHODS

Ambient HAP concentrations associated with wildfire smoke in the western US were characterized using two approaches: (1) a comprehensive analysis of HAPs measured at US EPA monitoring stations in western US states (Figure 1) from 2006 to 2020 and (2) a case study of temporal trends in gas-phase HAPs measured in San Jose, CA. Each of the data sets is described in detail.

Figure 1.

Air Quality System (AQS) monitor locations in the western US that provided data for this study, including the National Air Toxics Trends Station (NATTS) in San Jose, CA. For the monitor locations of each individual HAP, see Figure S1.

2.1. HAPs Associated with Wildfire Smoke in the Western US in 2006 to 2020.

2.1.1. Identification of Smoke-Impacted Air Quality Measurements.

We utilized daily average (24 h) measurements of 21 HAPs (1,3-butadiene, acetaldehyde, acrolein, arsenic, benzene, cadmium, carbon tetrachloride, chloroform, chromium, cobalt, dichloromethane, ethylene dichloride, formaldehyde, lead, manganese, mercury, nickel, tetrachloroethylene, trichloroethylene, cis-1,3-dichloropropene, and trans-1,3-dichloropropene) from US EPA AQS²⁴ monitoring locations from 2006 to 2020, limited to an April to December fire season¹² for all statistical tests. HAPs listed in AQS as core HAPs, VOCs, or PM_2.5 species were included.^24,26 Measurement methods were consistent throughout the study period, with the exception of acrolein,²⁷ for which the measurements using the updated method are labeled as “verified” and those using the retired method are labeled as “unverified”. PM_2.5 speciation monitors and National Air Toxics Trends Stations (NATTS) samples in one-in-three and one-in-six day schedules, respectively,²⁸ and HAPs in AQS, but outside of these networks, were included in calculations if they met the minimum data requirements described later. Data below the minimum detection limit were uncensored, in accordance with US EPA guidance,²⁹ and comprehensive information on measurement methods and quality assurance is available from the US EPA.^26,28,30,31 Monitoring locations were spread across the western US (Figure 1; Figure S1).

We classified each 24 h measurement as smoke-impacted or nonsmoke-impacted using smoke plume perimeters from the NOAA Hazard Mapping System (HMS).^32,33 The HMS incorporates imagery from multiple NOAA and NASA satellites and outputs the spatial extent of smoke plumes detected at or above the ground level in the atmospheric column, published daily after manual quality review by NOAA staff. Measurements were classified as smoke-impacted if an overhead smoke plume was detected over the respective monitoring location at any time during the day using a spatial join.³⁴ The HMS has been shown to best capture regional-scale trends in air quality³⁵ and may not detect some smoke events, due to the limitations of remotely sensed data (see discussion for more detail).

2.1.2. Permutation Test of Concentration Differences.

Station- and year-specific HAP concentrations on smoke vs nonsmoke days were compared to isolate the effect of smoke impact and control for variation across monitor locations and time. We conducted a nonparametric permutation test³⁶ to test whether concentrations were higher on smoke days compared with nonsmoke days (i.e., testing the null hypothesis that there were no differences) both with all years grouped together and each year individually from 2006 to 2020. This analysis was adapted from previously published methods.^12,37 This nonparametric test was selected for multiple reasons, including an unequal number of samples from smoke and nonsmoke days, and skewed concentration distributions. Observations were excluded in cases where the following conditions were not met at an individual monitor in a single year: (1) at least 20 measurements overall and (2) at least four measurements labeled as smoke-impacted. A detailed description of the permutation test methods is given in Section S1. Depending on the HAP, 20−67% of April−December observations were included (Table S1). An analysis of the permutation test’s sensitivity to seasonality was conducted, considering observations from May to September and in three month increments, in addition to the April-December fire seasons selected for the main analyses. Susceptibility to family-wise Type 1 error was evaluated by applying the Bonferroni correction (described in detail in Section S1).

2.1.3. Correlation with PM_2.5.

To identify HAPs co-occurring with PM_2.5 in wildfire smoke, we performed a correlational analysis using Pearson’s r to compare HAPs with PM_2.5 on smoke and nonsmoke days with collocated PM_2.5 federal reference method/federal equivalent method²⁴ measurements. We performed the analysis for each HAP for all days and on smoke and nonsmoke days separately.

2.2. San Jose Case Study, 2017 to 2020.

2.2.1. Major Wildfire Events Impacting the San Jose Air Monitoring Station from 2017 to 2020.

Beyond our overview analysis for the western US, we also conducted a case study using the San Jose NATTS site. The San Jose site is an urban air monitoring station that recorded data for the 2006 to 2020 study period and was impacted by smoke from multiple wildfires, including the selected case studies (the Tubbs, Camp, Kincade, and August Complex fires from 2017 to 2020).³⁸ Measurement methods were consistent throughout the time period for all pollutants, although the sampling schedule for gas-phase HAPs changed from one-in-six days to one-in-12 days and a different laboratory began analyzing the samples on July 1, 2018.³⁹ Acrolein was recorded by using the unverified sampling method. The site is located near downtown San Jose and the San Jose International Airport. The surrounding area has a population of nearly two million people, the largest population of any county in Northern California.²⁵

2.2.2. HYSPLIT Back-Trajectory Modeling of Major Fires from 2017 to 2020.

By visually inspecting time-series data, we identified peak concentrations from 2017 to 2020 for the HAPs with the greatest median concentration increases on smoke days identified in Table 1 (i.e., formaldehyde and acetaldehyde). To investigate the linkage between fires and these elevated concentrations, we employed NOAA’s Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) backward trajectory modeling.^40,41 HYSPLIT predicts air mass movement and was used to determine whether a smoke plume from a wildfire impacted the monitor. We calculated backward air mass trajectories starting at the San Jose NATTS station on the dates of the five 2017-to-2020 days with multiple elevated HAPs. We did this by modeling air mass trajectories every 3 h for 24 h (8 total traces) at 10 m above the ground level to identify if ground-level air masses passed over the site of the fires.^42,43 Representative 24−48 h air mass trajectories were overlaid with burn scars from the Monitoring Trends in Burn Severity program⁴⁴ during the time period of the fire to establish the probable source of the HAPs (Figure 2). Frequency heat maps were generated on the same dates by plotting the percentage of trajectories passing through each grid square (Figure S2). Using a larger number (8−16) of trajectories confirmed representative air mass transport.⁴⁵

Figure 2.

Map of Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) back-trajectory modeling of peak measurements at the San Jose National Air Toxics Trends Station (NATTS) air quality monitor. Modeling was performed for the Tubbs fire on October 16, 2017; the Camp fire on November 10, 2018; the Kincade fire on October 24, 2019; the beginning of the August Complex fires on August 19, 2020; and the Doe Fire during the longer August Complex event on September 12, 2020.

2.3. Comparison with Human Health RfC Values.

To place the potential health risk of HAPs from wildfire smoke into context, HAP concentrations were compared to RfC values for acute and chronic health effects used in risk assessment by US EPA.⁴⁶ Acute values included in this study represent the limit for no adverse health effects resulting from 1 h of exposure, and chronic values represent the maximum acceptable limit for a lifetime of exposure, unless noted otherwise. We considered daily measurements in comparison with acute RfCs and annual or multiyear mean concentrations in comparison with chronic RfCs. We additionally compared daily concentrations to the chronic RfC to characterize spatial and temporal trends in HAP concentrations, recognizing that an individual daily measurement greater than the chronic RfC does not generally indicate a concern for human health. This assessment may provide an indication of where chronic exposure could occur given an increase in wildfires.

Acute RfCs were from California EPA’s Reference Exposure Levels program.⁴⁶ Depending on the HAP, chronic noncancer RfCs were obtained from the Agency for Toxic Substances and Disease Registry,⁴⁷ California EPA,⁴⁸ and US EPA.⁴⁶ Human health RfCs for the HAPs studied are presented in Table 2. When calculating station-specific annual averages to compare to chronic RfCs, stations with fewer than four measurements in a year were excluded. Because of uncertainties associated with the unverified acrolein measurements, only measurements using the verified measurement method, beginning in 2011, were considered for this analysis.

Python 3.8 statistical data analysis and geographic information system packages were used for all analyses except for the permutation tests, which were performed by using R version 4.1.3. P values less than or equal to 0.05 were considered significant for all analyses.

3. RESULTS

3.1. HAPs Associated with Wildfire Smoke in the Western US from 2006 to 2020.

Of the 21 HAPs surveyed in our analysis, six were routinely elevated on days impacted by wildfire smoke. Acetaldehyde, acrolein, chloroform, formaldehyde, manganese, and tetrachloroethylene were significantly elevated on days with overhead smoke plumes in the western US compared with days without wildfire smoke, according to the permutation test across all years (from 2006 to 2020; Table 1). Benzene and 1,3-dichloropropene were not significantly elevated on smoke days across all years but were significantly higher in five or more years during the study period. The remaining HAPs, such as PM_2.5-size fraction lead, were significantly elevated in four or fewer years, and two (PM₁₀ size fraction lead and PM_2.5 cadmium) were not significantly elevated in any years (Table 1). Results for all 21 HAPs are presented comprehensively in the tables and figures of this article, while key results are highlighted in this section.

Table 1.

Absolute and Percent Differences and Annual Permutation Test Results for Hazardous Air Pollutants on Smoke- vs Nonsmoke-Impacted Days between 2006 and 2020 for the Western US

	Absolute difference (μg/m3)a				Percent differencea				Permutation test
Pollutant	Median	Mean	99th %ile	Max	Median	Mean	99th %ile	Max	P-value, overall	N years p < 0.05
Formaldehyde	1.3	1.4	7.0	18	46	55	190	490	0.000*	10
Acetaldehyde	0.73	0.84	4.3	9.8	36	53	310	330	0.000*	9
Acrolein-verified	0.14	0.16	0.89	0.98	34	52	260	270	0.000*	4
Benzene	0.095	−0.22	2.9	4.3	19	54	590	660	0.277	8
Acrolein-unverified	0.082	0.16	1.3	1.4	21	18	140	180	0.892	6
Carbon tetrachloride	0.0053	0.0012	0.14	0.24	0.82	−0.027	70	77	0.073	3
Manganese (PM2.5)	0.0006	0.0009	0.006	0.063	72	110	720	2800	0.000*	15
Lead (PM2.5)	0.00	0.00	0.0023	0.0064	6.7	36	520	4100	0.709	2
Lead (PM10)	0.00	0.0002	0.0028	0.0031	0.76	7.6	140	170	0.191	0
Chloroform	0.00	0.024	0.30	0.59	15	85	1300	2100	0.000*	5
Ethylene dichloride	0.00	−0.0011	0.054	0.089	−9.5	−19	220	280	1.00	2
Arsenic (PM2.5)	0.00	0.00	0.0006	0.0043	−100	6.2	1200	2300	1.00	2
Chromium (PM2.5)	0.00	−0.0002	0.0028	0.016	−55	13	1100	4600	0.766	1
Cis-1,3-dichloropropene	0.00	−0.013	0.038	0.043	−8.5	−14	380	390	0.985	4
Tetrachloroethylene	0.00	0.075	1.0	7.3	−11	25	720	1200	0.019*	5
Trichloroethylene	0.00	0.0094	0.22	0.91	−0.47	80	1400	2200	0.459	4
Nickel (PM2.5)	0.00	−0.0001	0.0014	0.0046	−23	42	1000	4600	0.305	1
Trans-1,3-dichloropropene	0.00	−0.0079	0.025	0.029	0.00	−0.064	350	390	0.953	5
Mercury (PM2.5)	0.00	0.00	0.0011	0.0017	0.00	−11	230	270	0.919	2
Cobalt (PM2.5)	0.00	−0.0004	0.0009	0.0016	−12	−6.1	280	350	0.999	2
Cadmium (PM2.5)	−0.0003	−0.0001	0.0054	0.0062	−16	9.4	400	1200	0.697	0
1,3-butadiene	−0.0006	0.0034	0.24	0.51	−19	66	1000	6100	0.762	3
Dichloromethane	−0.025	−3.2	21.0	290	−8.4	27	970	1300	0.580	2

^aDifferences in pollutant concentrations on smoke-impacted days compared with nonsmoke days, ordered by the median concentration difference. To calculate differences in smoke-impacted days, year and station-specific mean differences between days impacted by smoke or not were calculated for each monitor in each year with at least four smoke-impacted days and 20 measurements overall. Median, mean, and max differences in both absolute and percentage terms were calculated from the distribution of year and station-specific differences. Results from the permutation test are reported as overall P values and as a count of individual years with statistically significant differences. %ile = percentile;

^*P < 0.05.

The magnitude of concentration increases on smoke-impacted days varied among the HAPs, in both absolute and percentage terms (Figure 3). The HAPs with the greatest concentration increases on smoke days compared with nonsmoke days were formaldehyde (median increase: 1.3 μg/m³ [46%]; maximum increase: 18 μg/m³ [490%]) and acetaldehyde (median increase: 0.73 μg/m³ [36%]; maximum increase: 9.8 μg/m³ [330%]) (Table 1; Figure 3). Some HAPs studied had small increases in absolute concentrations, but large percent increases compared with days without wildfire smoke. For instance, the median difference for PM_2.5 size-fraction manganese was less than 0.01 μg/m³ on smoke vs nonsmoke days, yet this represented a 72% increase. Table 1 comprehensively presents concentration differences between smoke and nonsmoke days, Table S1 presents summary statistics, and Table S2 shows observation counts by year.

Figure 3.

Box-and-whisker plots of absolute and percent mean differences between smoke-impacted and nonsmoke-impacted days at each included US EPA Air Quality System monitoring station per year. The green-shaded boxes show the 25th, 50th, and 75th percentiles of the distribution of station- and year-specific differences, and whiskers extend 1.5 times the interquartile (25th−75th percentile) range.

In an analysis of the permutation test’s sensitivity to the fire season definition, equivalent or stronger results were obtained by analyzing a shortened May to September fire season for each of the six pollutants identified as elevated on smoke days in the April−December fire season (Table S3). Four of the six identified pollutants were significantly elevated or near significant (i.e., P < 0.10) in the permutation test for each of the three month periods (April−June, July−September, and October−December). Tetrachloroethylene was significantly elevated in April−June and July−September tests but not in the October−December period. Sufficient data were lacking to conduct the permutation test for verified acrolein in April−June and October−December, yet the test was significant for the July−September period when enough data were available. Beyond these six HAPs, several additional chemicals were significantly elevated depending upon the time period analyzed (Table S3). These were carbon tetrachloride, benzene, trichloroethylene, lead (PM_2.5), 1,3-butadiene, and trans- and cis-1,3-dichloropropene. The remaining HAPs were not significantly elevated across any monthly time increment. Investigating the permutation test’s susceptibility to family-wise Type 1 error using the Bonferroni correction, the results were consistent with those reported in Table 1 except for the overall P value for tetrachloroethylene, which was no longer significant at the alpha = 0.0002 level, and the individual year tests for benzene, which were significant for 3 rather than 4 years (Table S4).

Higher HAP concentrations occurred more often on smoke-impacted days than on nonsmoke days. Among days with the highest concentrations of HAPs (above the 90th percentile), a larger proportion of measurements were smoke-impacted than those in lower percentiles of the HAP distributions. Formaldehyde, manganese, acetaldehyde, acrolein, benzene, carbon tetrachloride, chromium, and chloroform all had greater proportions of smoke-to-nonsmoke measurements in the top 10th percentiles compared with below the 90th percentile (listed in order of descending percentage-point difference; Table S5). Absolute maximum values of most HAPs, however, were observed on days without an overhead smoke plume (Table 2, Table S5), with a few exceptions (e.g., acrolein and tetrachloroethylene).

Additionally, HAPs associated with smoke were measured at higher concentrations in 2017, 2018, and 2020 compared with earlier years of the study period. In the permutation test performed for each year in the study, associations were generally stronger in later years (Figure 4). In 2018 and 2020, 15 and 12 of the HAPs, respectively, were significantly higher (P < 0.05) on smoke-impacted days, whereas in the other years of the study, the number of HAPs higher on smoke-impacted days ranged between 1 and 9.

Figure 4.

Permutation test results for concentration differences on smoke-impacted days by year (P values). Compounds are sorted in descending order by the P value of the permutation test for all years overall. Blank cell values indicate the minimum data requirements for the year-specific permutation test were not met.

Some of the elevated HAPs were moderately (r > 0.4; acetaldehyde, 1,3-butadiene) or strongly (r > 0.7; benzene, acrolein) correlated with total PM_2.5 on smoke days. HAPs including benzene, acrolein, acetaldehyde, 1,3-butadiene, formaldehyde, carbon tetrachloride, trichloroethylene, and cadmium (listed in order of descending smoke day correlation with PM_2.5) had higher correlations with PM_2.5 on days impacted by smoke plumes compared with nonsmoke days throughout the study period (Figure S3). Because PM_2.5 is a pollutant typically elevated in smoke,³⁵ stronger correlations on smoke days relative to nonsmoke days suggest those HAPs are likely elevated in smoke.

3.2. San Jose Case Study, 2017 to 2020.

Peak concentrations of HAPs in San Jose from 2017 to 2020 were linked to four major fires using five HYSPLIT backward trajectories: the Tubbs fire of 2017, the Camp fire of 2018, the Kincade fire of 2019, and the August Complex fires of 2020, which included the Doe fire.³⁸ All backward air mass trajectories passed over or near an associated fire boundary (Figure 2). Concentrations of select HAPs are shown in Figure 5, while Figure S4a,b shows each HAP measured in San Jose. Among HAPs identified as significantly enhanced on smoke days in the permutation test across all western US monitors, the levels of acetaldehyde, acrolein, and formaldehyde were most consistently elevated on days linked to the four wildfire events. Notably, the two highest formaldehyde measurements in 2006 to 2020 in San Jose occurred during the Camp fire in 2018 (11.2 μg/m³) and the August Complex fire in 2020 (11.4 μg/m³), despite a reduction in sampling frequency to one-in-12 days in those years. Of the chlorinated species, chloroform was the only one observed elevated during the wildfire events (Figure 5). Concentrations of benzene were highest during the case study fire events (Figure S4a), although the species was not identified as elevated in the permutation test applied to measurements from April to December. Benzene was, however, identified as significantly enhanced on smoke days in the permutation test analyzing a shorter May to September fire season (Table S3). Elevated manganese was observed during the fires, but maximum concentrations were below 0.02 μg/m³ (Figure S4b).

Figure 5.

Time-series of select hazardous air pollutants measured in San Jose from 2017 to 2020. The dark gray vertical annotations represent the day back-trajectory modeling was used to identify smoke influence, and the lighter gray shading represents the overall duration of the Tubbs, Camp, Kincade, and August Complex (including Doe) fires. To improve the legibility of this figure, pollutants with higher absolute concentrations are plotted on the upper plot, and pollutants with lower concentrations are plotted on the lower plot. Measurements of acrolein at San Jose were obtained using the unverified method with additional uncertainty compared with the verified method.

3.3. Comparison with Human Health RfC Values.

We found certain instances where HAP concentrations on smoke days in the western US were greater than RfCs for acute health effects, thresholds above which 1 h of exposure may cause adverse health effects (Table 2). In these cases, the 24 h average measurement exceeded the 1-h RfC for acute health effects of acrolein, benzene, and formaldehyde. This is a conservative metric because the 1 h RfC could be exceeded for multiple hours in the 24 h period. Across all monitors in the western US during the 2006−2020 study period, acrolein was measured greater than the acute California EPA reference exposure level (REL) on smoke-impacted days six times; benzene exceeded the threshold twice; and formaldehyde exceeded the threshold once on a smoke-impacted day. Only acrolein was found to have mean concentrations across the study duration that exceeded the chronic RfC (Table 2). Mean concentrations across all smoke and nonsmoke days for acrolein exceed the RfC of 0.35 μg/m³, with higher concentrations on smoke vs nonsmoke days (0.65 and 0.55 μg/m³, respectively, averaged across all sites). Among the HAPs analyzed during major fires impacting San Jose, daily HAP measurements did not exceed acute RfCs (with the exception of unverified acrolein, which we did not compare to RfCs due to methodological uncertainty).

Table 2.

Mean and Maximum Smoke Day Concentration, Acute and Chronic Human Health Reference Concentrations (RfCs), and Number of Exceedances of Those Values for Hazardous Air Pollutants across All Western US Monitors from 2006 to 2020^e

					Acute RfCs				Noncancer chronic RfCs
							Count of days exceeding				Count of days exceeding
Chemical name	Smoke day mean (μg/m3)	Smoke day max (μg/m3)	Nonsmoke day mean (μg/m3)	Nonsmoke day max (μg/m3)	RfC (μg/m3)	RfC source	Smoke	Nonsmoke	RfC (μg/m3)	RfC source	Smoke	Nonsmoke
1,3-butadiene	0.06	3.3	0.08	11	660	REL	0	0	2	IRIS	2	9
1,3-dichloropropenea	0.12	3.2	0.13	12			0	0	20	IRIS	0	0
Acetaldehyde	3.3	32	2.5	43	470	REL	0	0	9	IRIS	87	360
Acroleinb	0.65*	5.1	0.55*	4.6	2.5	REL	6	9	0.35	CAL	213	1335
Arsenic compoundsc	0.00	0.02	0.00	0.04	0.2	REL	0	0	0.015	CAL	2	19
Benzene	0.80	40	0.88	4800	27	REL	2	2	30	IRIS	1	2
Cadmium compoundsc	0.00	0.03	0.00	0.05			0	0	0.01	ATSDR	205	2242
Carbon tetrachloride	0.58	3.3	0.58	3.20	1900	REL	0	0	100	IRIS	0	0
Chloroform	0.12	2.7	0.13	11	150	REL	0	0	98	ATSDR	0	0
Cobalt compoundsc	0.00	0.01	0.00	4.0			0	0	0.1	ATSDR	0	1
Ethylene dichloride	0.02	1.4	0.03	1.3			0	0	2400	ATSDR	0	0
Formaldehyde	4.9	56	3.7	91	55	REL	1	1	9.8	ATSDR	147	674
Lead compoundsc	0.00	0.10	0.00	0.98			0	0	0.15	NAAQSd	0	22
Manganese compoundsc	0.00	0.31	0.00	0.73			0	0	0.3	ATSDR	1	4
Mercuryc	0.00	0.01	0.00	0.02	0.6	REL	0	0	0.3	IRIS	0	0
Methylene chloride	3.7	920	5.4	11000	14000	REL	0	0	600	IRIS	2	18
Nickel compoundsc	0.00	0.08	0.00	3.0	0.2	REL	0	3	0.09	ATSDR	0	14
Tetrachloroethene	0.14	50	0.13	22	20000	REL	0	0	40	IRIS	1	0
Trichloroethylene	0.04	3.8	0.10	260			0	0	2	IRIS	3	65

^aMean and maximum cis-1,3-dichloropropene values are reported in this table. Cis- and trans-isomers are measured separately in the AQS network, but the RfC is not isomer-specific.

^bFor acrolein, only samples using the verified method are included in this table. IRIS = Integrated Risk Information System. ATSDR = Agency for Toxic Substances and Disease Registry. NAAQS= National Ambient Air Quality Standard. REL = California EPA’s reference exposure level. RfC = reference concentration (for inhalation). CAL = California EPA.

^cFor all metals and arsenic, only PM_2.5 size fraction concentrations were measured at AQS sites; total suspended particle concentrations may exceed the reported maximum values.

^dThe chronic RfC for lead is the US EPA NAAQS, with a three month averaging time. In cases where the RfC is left blank, no data were available from the cited sources.

^eAcute RfCs refer to the maximum concentration for no adverse effects for 1 h of exposure, and chronic RfCs refer to a concentration below which no effects are expected for a lifetime of exposure. Mean and maximum concentrations were calculated across all monitors in 2006−2015 in April−December.

^*Mean concentration exceeding the chronic RfC.

The proportion of daily measurements greater than both acute and chronic RfCs occurring on smoke-impacted days generally increased in later years of the study period (i.e., 2017, 2018, and 2020; Figures 6 and 7). In 2017 and 2020, mean smoke day acrolein concentrations were greater than the chronic RfC at more stations than nonsmoke day mean concentrations (although, verified acrolein measurements, analyzed here, began in 2011 and the number of measurements increased year over year; see Table S2a,b). In 2020, HAP measurements exceeding the acute RfC occurred solely for acrolein and benzene on smoke-impacted days. Daily concentrations greater than chronic RfCs occurred on smoke days more often in 2018 (29%) and 2020 (40%) than in other years (4−19%; Figure S5). Besides generally increasing in later years, daily concentrations greater than chronic RfCs on smoke days tended to cluster spatially (Figure 7). As depicted in Figure 7, some air quality monitors, particularly in California and Oregon, experienced significant increases in HAP concentrations that could indicate a trend toward the surrounding areas potentially being chronically exposed to HAP concentrations that are of health concern.

Figure 6.

Counts of measurements greater than hazardous air pollutant reference concentrations (RfCs) on smoke and nonsmoke days by year. Daily measurements were considered in comparison with acute RfCs, and both daily and annual average concentrations at individual stations were considered in comparison with chronic RfCs. Verified measurements of acrolein were not available until 2011 and the number of observations increased through 2020. Conc. = concentration.

Figure 7.

Percent of daily measurements greater than chronic hazardous air pollutant reference concentrations occurring on smoke days by monitoring location. Individual panels show time periods of 2006−2009, 2010−2013, 2014−2016, and 2017−2020.

4. DISCUSSION

Overall, concentrations of six HAPs (acetaldehyde, acrolein, chloroform, formaldehyde, manganese, and tetrachloroethylene) were significantly elevated on smoke days in the western US from 2006 to 2020 (Table 1). Of these HAPs, formaldehyde had the largest concentration difference between wildfire smoke days and nonsmoke days, which is potentially significant for human health since formaldehyde contributes the greatest national-scale cancer risk of the HAPs.²¹ Two other HAPs, benzene and 1,3-dichloropropene, were also linked to smoke in more than four of the years studied. Due to the spatial and temporal variability in smoke, there was high variability in HAP concentrations between monitors and years. Wildfires are highly stochastic events, varying in extent, severity, and types of fuels burned, among other characteristics.^11,20 Despite this variability, the long-term data used here enabled the characterization of HAPs over time and an entire region at ground-level monitoring stations located largely in the population centers. To our knowledge, this study is the first to examine the link between concentrations of HAPs and wildfire smoke using multiyear air quality network monitoring data.

This study of air quality monitoring data builds on a long history of research on emissions from biomass burning, which has identified a wide range of chemical species emitted from wildfires, in addition to the major combustion products (CO₂, CO, CH₄, and PM_2.5).²⁰ Apart from the major combustion products, wildfire emissions include organic species in the gas and particle phases, inorganic gases (e.g., HCl, NOx, NH₃), and inorganic species in the particle phase (e.g., sulfate, trace metals).²⁰ Numerous laboratory studies and field studies have quantified emission factors with much of the focus on VOCs since they are a major portion of the emissions and are important precursors for ozone and secondary organic aerosol formation. Only recently has there been an increasing focus on HAPs emitted from wildfires.^10,12

In this study, five of the six HAPs consistently elevated on smoke-impacted days can be divided by their chemical properties into two groups: oxygenated and chlorinated species. The oxygenated species (formaldehyde, acetaldehyde, and acrolein) are emitted in the largest amounts by wildfire²⁰ and are some of the major products from the pyrolysis of biomass, the dominant process in smoldering combustion. Overall, oxygenated species comprise nearly 70% of the measured VOCs in wildfire plumes.⁴⁹ Elevated formaldehyde and acetaldehyde concentrations have been observed in populated urban areas impacted by wildfire activity,^50–52 and biomass burning has been described as the predominant source of atmospheric acrolein.⁵³ Chlorinated hydrocarbons (chloroform and tetrachloroethylene) from wildfires have also been studied^54–56 and reported in measurements in fire plumes and emissions studies.^10,11,57 O’Dell et al.¹⁰ reported chloroform and tetrachloroethylene in wildfire smoke plumes in the western US but did not measure carbon tetrachloride. In contrast, Aurell et al.⁵⁷ were unable to determine an emission factor for chloroform or tetrachloroethylene from laboratory forest burns but did identify carbon tetrachloride.

A few HAPs not identified in wildfire smoke in this study did have emissions factors reported,¹¹ including 1,3-butadiene, benzene, carbon tetrachloride, dichloromethane, and ethylene dichloride. For these HAPs, it is possible that seasonality influenced our ability to detect enhancement in wildfire smoke. Both 1,3-butadiene and benzene typically have higher concentrations in winter, in the absence of smoke,^58,59 and were identified in the permutation test as significant with a shorter May-September fire season (Table S3), suggesting that higher nonsmoke values later in the fire season (e.g., November/December) could have masked differences. It is also possible that concentrations in smoke were at times too small to detect compared with other sources (e.g., carbon tetrachloride),^57,60 and our method was unable to detect pollutants with short atmospheric lifetimes (e.g., benzene).¹⁰ Emissions of chlorinated compounds are more likely in smoldering burns and from vegetation in coastal areas,⁶¹ so burning conditions and distance from the ocean could influence concentrations.

Notably, there is a possibility of enhanced emissions for certain oxygenated and chlorinated species from WUI fires, where combustion of building materials, furniture, household appliances, and vehicles can occur. The oxygenated species have all been observed in emissions from structure and car fires.^62–67 A recent National Academy of Sciences report¹⁸ identified HAPs, including ones observed here (formaldehyde, benzene, and acrolein) as potential primary emissions from WUI fires, and emissions of chlorinated hydrocarbons have recently been documented.⁶⁷ Although we are unable to exclusively link WUI sources to the concentrations observed in this study, the highest values of formaldehyde and acetaldehyde measured between 2006 and 2020 in San Jose were observed from air masses traveling over the Camp fire in 2018, which burned more structures than any other fire in California’s history.³⁸ Local maximum chloroform and benzene concentrations were also observed (Figure S4a).

As with other air pollutants, fires are countering the benefits of reductions of HAPs in other sectors.⁶⁸ Concentrations of PM_2.5 in the US have been declining for decades because of air quality management practices,⁶⁹ but fires now account for an estimated 50% of total PM_2.5 concentrations in some areas of the western US.¹³ Similarly, Jerrett et al.⁷⁰ estimated increases in greenhouse gases from fires in 2020 likely more than canceled out the greenhouse gas emissions reductions achieved by California from 2003 to 2019. In California, formaldehyde and acetaldehyde concentrations decreased by 22 and 21% from 1996 to 2012, respectively, and ambient concentrations of tetrachloroethylene decreased by 92% during the same period.⁷¹ Nationwide decreases from 2003 to 2013 of 17% for formaldehyde, 28% for acetaldehyde, and 77% for tetrachloroethylene observed by US EPA were similar.⁷² Our analysis of ambient concentrations exceeding RfCs shows that fires are, in part, undermining these trends, with 29% of daily measurements in excess of chronic RfCs occurring on smoke-impacted days in 2018 and 40% in 2020, compared to a range of 4−19% earlier in the study period (Figure S5).

The observation that wildfires are causing significant increases in certain HAPs during smoke-impacted days raises potential health concerns, in addition to those well documented from PM_2.5 exposure from wildfire smoke.¹³ To assess whether the increase in HAP concentrations reported in this analysis is of health concern, they were compared with the current acute and chronic human health RfCs. With respect to the acute RfCs, there were relatively few instances of exceedance across the HAPs evaluated. From a population-level perspective, this equates to a limited health concern, but first responders and wildland firefighters working close to the fire, although not the focus of this study, may be exposed to higher HAP concentrations than those measured at the monitoring sites evaluated, often far-downwind of a fire epicenter.^73,74

When comparing average concentrations to the chronic RfC, only acrolein was found to exceed the chronic RfC, and it did so on both smoke and nonsmoke days. Widespread exceedance of RfC values for acrolein is consistent with previous observations.^75,76 Although measurements averaged over longer time periods are needed to assess whether longer-term exposures to HAPs during wildfire events are of health concern, daily average concentrations greater than the chronic RfC were evaluated to investigate spatial and temporal patterns in HAPs concentrations during smoke-impacted days. Acetaldehyde, acrolein, cadmium compounds, and formaldehyde had the most daily concentrations above chronic RfC, ranging from 87 to 213 smoke days over the study duration. The number of smoke days where the average HAP concentration exceeded the chronic RfC equates to a small percentage of total smoke days for which these HAPs were measured (i.e., 4−7% for acetaldehyde, cadmium, and formaldehyde; and 63% for acrolein but based on fewer measurements). However, as wildfire activity (i.e., area burned, total number of fires, and length of the fire season)^3–5 is projected to increase in response to climate change, the number of days with average HAPs concentrations above the chronic RfC could increase. This could ultimately shift the concentration distribution for some HAPs, resulting in an exceedance of the chronic RfC during smoke-impacted days.^4,77

Importantly, the HAP concentrations observed in this study represent the combined ground-level concentrations of HAPs from wildfire smoke as well as urban and industrial sources. This additional burden of HAPs from wildfire smoke could alter disparities in HAPs exposure across communities. While there are well-defined disparities in exposure to ambient PM_2.5 in communities with environmental justice concerns⁷⁸ that could be further exacerbated by wildfire smoke, intraurban patterns of HAPs exposure are less well-defined. Future efforts are needed to assess whether increases in HAPs attributed to wildfire smoke could further increase overall HAP exposures in vulnerable communities.

To protect public health from wildfire smoke, current recommendations focus on exposure reduction and mitigation actions for PM_2.5 during smoke events, as a result of the clear relationship between PM_2.5 exposure and health effects.^78,79 The actions recommended include using high-efficiency particulate air filters and other indoor air cleaners,^80,81 but it is unclear how effective they are at reducing other pollutants, such as some of the gaseous HAPs observed here, in wildfire smoke broadly or elevated due to the burning of human-made materials and structures.^15,18,82 Air cleaners are not routinely certified for VOC removal, although standardized methods now exist.⁸³ Studies have found variable VOC removal rates, depending on cleaning technology⁸⁴ with some technologies adding to the pollution by emitting oxygenated hydrocarbon byproducts.⁸⁵ As area burned and the frequency of large wildfires continues to increase along with the probability of these fires entering the WUI,^3–5,18 it will become even more important to understand the capabilities of exposure reduction and mitigation options in reducing HAPs exposures.

As in all studies, the interpretation of the results here is subject to several limitations. Ambient concentrations of 21 of 188 HAPs were assessed in this study overall. Hydrogen cyanide, dioxin, and polycyclic aromatic hydrocarbons are examples of HAPs not included in this study that may contribute to the health impacts of wildfire smoke.²⁰ In the HAPs reported, infrequent measurements (typically a 3 day sampling schedule for PM_2.5 components, 6 days for gas-phase HAPs, or 12 days in San Jose after July 1, 2018) prevented us from assessing the exposure duration experienced by downwind populations. Given more frequent sampling, it is possible that higher magnitude values would have been recorded. Additionally, changes in measurement methods during the study period may have affected the observed concentrations. This was documented on a national scale for acrolein²⁷ but may also apply to other HAPs, such as in San Jose, where the sampling frequency and analytical laboratory changed on July 1, 2018. The spatial distribution of monitors is not uniform across the western US, and it varies by year and by HAP (Figure S1), which likely introduces some year-to-year and between-HAP heterogeneity in the results of this work. Finally, we did not analyze sources other than wildfires in this study. By comparing station- and year-specific concentration differences; however, to the extent possible, we controlled for variation in nonwildfire sources and measurement methods across monitor locations and over time. Likewise, in the case study, we are unable to account for changes in HAP sources aside from the selected fires; the reported concentrations represent the total, regardless of source.

In addition to limitations in the measurement of HAPs, we relied on remote sensing data to identify smoke-impacted days, specifically the NOAA’s HMS data product. The HMS may misclassify days that are truly smoke-impacted as not smoke-impacted (e.g., if a cloud obfuscates the smoke, the event occurs at night, or the smoke is too thin to be identified) or misclassify days that are in fact not smoke-impacted as smoke-impacted (e.g., if the smoke plume occurs far above the ground monitor in the atmospheric column); however, days defined as smoke-impacted have been found to report consistent increases in air pollutant concentrations, specifically PM_2.5.³⁵ It is unclear if the potential misclassification of smoke and nonsmoke days using HMS would lead to an underestimate or overestimate of HAPs.

Despite these limitations, this study shows that wildfire smoke increased the concentrations of HAPs observed in downwind communities across the western US between 2006 and 2020. In particular, six HAPs were significantly elevated on smoke-impacted days: acetaldehyde, acrolein, chloroform, formaldehyde, manganese, and tetrachloroethylene. These findings have implications for acute and long-term health effects. Exceedances of the acute and chronic human health RfCs for HAPs occurred more often on smoke-impacted days in recent years, particularly in 2018 and 2020, consistent with increasing fire activity.^3,4 It is unclear whether current filtration methods can reduce exposure to gas-phase HAPs.^15,18,82 This problem is not expected to lessen with time, given current trends in wildfire coupled with likely increases in the future due to climate change.^6,86 By characterizing these HAPs in wildfire smoke, we hope to improve our understanding of the risk to public health and ultimately aid in efforts to reduce the exposure to these chemicals.

Data Availability Statement

The data and code underlying this study are openly available in Data.gov at doi.org/10.23719/1529861.