Evolution of reactive organic compounds and their potential health risk in wildfire smoke

Abstract

Wildfires are an increasing source of emissions into the air, with health effects modulated by the abundance and toxicity of individual species. In this work, we estimate reactive organic compounds (ROC) in western U.S. wildland forest fire smoke using a combination of observations from the 2019 Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) field campaign and predictions from the Community Multiscale Air Quality (CMAQ) model. Standard emission inventory methods capture 40–45% of the estimated ROC mass emitted with estimates of primary organic aerosol particularly low (5–8×). Downwind, gas-phase species abundances in molar units reflect the production of fragmentation products such as formaldehyde and methanol. Mass-based units emphasize larger compounds which tend to be unidentified at an individual species level, are less volatile, and are typically not measured in the gas phase. Fire emissions are estimated to total 1250 ± 60 g C of ROC per kg C of CO, implying as much carbon is emitted as ROC as is emitted as CO. Particulate ROC has the potential to dominate the cancer and noncancer risk of long-term exposure to inhaled smoke, and better constraining these estimates will require information on the toxicity of particulate ROC from forest fires.

Graphical Abstract

Introduction

Wildfire activity has been increasing in response to changes in climate¹ with implications for the earth’s radiative forcing² and human³ and ecosystem⁴ health. Globally, exposure to air pollution from wildfires has been estimated to result in 339,000 deaths a year.³ Within the United States, premature deaths from long-term exposure to fire smoke are reported to result in a loss of roughly $100 billion dollars a year in economic value (2010 dollars).⁵ In some locations in the United States, increases in wildland fire activity have already caused a stagnation or reversal of multi-decadal improvements in air quality,⁶ and climate change is predicted to lead to further increases in wildland fire activity and its associated air pollutants.^{7, 8}

Wildfire smoke is a complex mixture of pollutants, and individual components can contribute to a range of health effects. For example, woodsmoke contains at least 26 chemicals classified as hazardous air pollutants (HAPs) by the U.S. Environmental Protection Agency (EPA) and 5 chemical groups that are known human carcinogens according to the International Agency for Research on Cancer (IARC).⁹ Characterizing and controlling for exposure to different individual smoke components in an ambient air study is challenging, and particulate matter (PM) mass often serves as the best indicator of exposure to wildfire smoke. Thus, PM is commonly used to estimate the health effects of smoke,^{3, 9–11} and limited studies have examined PM and HAPs together.^{12, 13} As a result, the role of gas- vs. particle-phase species on health impacts from fire smoke has not been robustly characterized. A better understanding of the role of gases vs. particles in driving adverse health outcomes can improve strategies to reduce the harmful effects of smoke as filtration, such as N95 masking and the use of fibrous filters, is effective for removing particles but not gases.^{14, 15} Proper selection of exposure mitigation strategies can also avoid unnecessary technologies and harmful byproducts.^{14, 16, 17}

Organic compounds account for most of the emitted fine particulate matter (PM_2.5, particles with an aerodynamic diameter ≤2.5 μm) mass from wildfires,^18–21 and gas-phase organic emissions can be of similar abundance as the particulate organic emissions from fires.^{22, 23} The combined gas and particle pool of reactive organic compounds (ROC), which includes all organic species except methane, drives atmospheric reactivity and formation of secondary products²⁴ and contains HAPs.¹³ ROC emissions evolve as they travel downwind from fires, and even for species that appear roughly constant relative to CO, as in the case of formaldehyde, chemical losses and secondary production simultaneously occur.^{25, 26} The particle portion of ROC has been widely demonstrated to be semivolatile with concentrations in downwind smoke modulated by dilution, evaporation, and chemistry.^27–31 However, as much as half of the gas-phase ROC from wildfires may not be identified at the individual species level and models generally do not have representative species for all identified emissions which challenges the ability of models to represent emissions and their transformation products.^32–35

In this work, we aimed to build a complete description of ROC emissions and their near-source gas vs. particle abundance in wildland fire smoke using a combination of observations and model predictions. Rather than the traditional terms organic aerosol (OA) and non-methane organic gases (NMOG), ROC_P and ROC_G refer to the particulate and gas forms of ROC, including the carbon and noncarbon mass, to emphasize that the two forms are connected and can interconvert. We define ROC as the set of reactive organic compounds to emphasize the inclusion of both carbon and noncarbon mass. Observations of ROC species are gathered from the DC-8 aircraft for western U.S. wildfires during the 2019 Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) field campaign³⁶ and compared to predictions from the Community Multiscale Air Quality (CMAQ) model using the Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMM) with AMORE isoprene chemistry.^{37, 38} After updating the emission inputs and chemical evolution of wildfire smoke to understand species abundance, concentrations are extended to estimates of cancer and noncancer risk that would arise from long-term exposure to ROC.

Methods

Wildfire Episodes

CMAQv5.4 simulations with CRACMM1AMORE chemistry,^{37, 38} hereafter referred to as CMAQ-CRACMM, were conducted from 1 July to 6 September 2019 (first 3 weeks used for initialization and not used in analysis) at 12 km by 12 km horizontal resolution with 35 vertical layers across the conterminous U.S. CRACMM includes a full suite of ROC and ozone chemistry including OA from oxidation of alkanes, oxygenated species (peroxides, aldehydes, etc.), aromatic species, furanone, phenol, cresols, sesquiterpenes, monoterpenes, and isoprene products.³⁹ Initial and boundary conditions were the same as previous work^{40, 41} and obtained from simulations with Carbon Bond chemistry (cb6r3_ae7) in the EPA’s Air QUAlity TimE Series (EQUATES) project.⁴² Gas-phase organic emissions followed standard methods in EQUATES⁴² speciated for CRACMM³⁹ with fire emission speciation a function of fire type (prescribed or wildfire) and geographic region⁴³ (Table S1–S2). In addition to the standard CRACMM1AMORE species, acrylonitrile, acetonitrile, and hydrogen cyanide (HCN) were added as tracers subject to deposition in the CMAQ simulation. Speciation of particle emissions was the same for all fires (Table S3). The inventory estimate of primary OA (the sum of organic carbon and the accompanying noncarbon organic matter) was used as an estimate of the total pool of ROC emissions of saturation concentration (C*) 10,000 μg m⁻³ and below (EF_10,000) with the volatility specified by May et al.²⁷ (Table S4).

Observations of ROC in both the gas and particle phases were obtained from FIREX-AQ measurements taken aboard the DC-8 aircraft³⁶ with techniques the same as those described by Gkatzelis et al.²² (Table S5). ROC_P observations and their organic matter to organic carbon ratio were obtained from a high-resolution time-of-flight aerosol mass spectrometer (AMS).^{30, 44} Custom data merges⁴⁵ at 1 Hz were averaged up to 60 s while ignoring all missing data. Organic carbon (OC) in ROC_P was calculated at the original AMS time resolution and then aggregated up to 60s. CMAQ instantaneous hourly concentration output was sampled coincident with the flight track in time (matching to the nearest hourly model output value) and space (using the radar altitude from the aircraft, model layer height, latitude, and longitude). Observational data were further subset based on the modified combustion efficiency (MCE = [CO₂]/([CO₂]+[CO]))^{46, 47} to remove smoldering-dominated conditions (MCE < 0.85 removed) and to emphasize forest fires west of 100°W longitude resulting in up to 792 possible data points for a given species. For model predictions, a threshold of 10 ppt predicted HCN from fires was further implemented to avoid including model output where fire activity was not predicted (476 data points). Criteria for data selection are summarized in Table S6 while Figures S1–S2 show the spatial and vertical distribution of data.

Concentration Analysis

Previous work found gas-phase ROC species in fresh smoke during FIREX-AQ correlated well with carbon monoxide (CO) indicating CO could be used as a proxy for emissions.²² In addition, CO was found to be a robust indicator of ROC_P abundance as ROC_P near fires during FIREX-AQ was driven by semivolatile partitioning.³⁰ To minimize the impact of slight spatial displacements of the fire plume within CMAQ-CRACMM compared to observations and to utilize CO as a proxy for emitted vs. downwind smoke, data were organized along a CO dimension. Species concentrations as well as CO were background corrected using the minimum value in smoke. The ratio of a species abundance to CO was then used to remove the first-order effects of dilution on concentration. The average, background-corrected species concentration relative to background-corrected CO (normalized excess mixing ratio) over the highest (97^th percentile and above) CO concentrations in smoke were used to calculate an inferred emission ratio (ER) in both CMAQ-CRACMM predictions and the FIREX-AQ observations. The abundance of total ROC_G in smoke was calculated as a sum of individual species. Downwind concentrations in smoke for individual species were determined by taking the median enhancement ratio for each decile in CO. Those trends were then interpolated across a range of CO in smoke and summed when totals were needed. See Figure S3 for an illustration of these steps. Note calculations for the lowest decile of CO can be sensitive to the background values and may not be as robust. In addition, while low CO conditions are categorized as downwind, the relationship between CO and age is complex and dependent on multiple factors such as fire size, meteorology, and burn conditions.

Species ratios versus CO were expressed in three different units: molar, mass, and mass of carbon. For gases, measurements were reported in mixing ratio units which leads to molar ratios (or ppb/ppm) when normalized by CO. For particles, measured concentrations are in mass and mass of carbon. Since a representative molecular weight is not readily available for particle measurements, total ROC was expressed in mass and mass of carbon units. Carbon mass is more likely than mass to be conserved in smoke as functionalization chemistry adds mass to the carbon backbone.

After examining the base CMAQ-CRACMM predictions for wildfire smoke, several species’ emission rates were updated to better reflect FIREX-AQ observations and produce a second CMAQ-CRACMM sensitivity simulation. Species missing from the base inventory were set to values from literature (cresols and similar (CSL),²² furanones (FURANONE)⁴⁸) or those inferred from this work at high CO in smoke (hydroxyketones (HKET), methylcatechols (MCT), methylglyoxal and similar structures (MGLY)). In cases where the CMAQ-CRACMM inferred ER was lower than observation-based ER by ≥50% and downwind trends suggested emission errors, species were scaled to better match the FIREX-AQ emissions inferred in this work. Phenol (PHEN) emissions were scaled down to recast the species as an individual compound rather than lumped category in the model. Specific species changes were verified using data from Akagi et al.,¹⁸ Koss et al.,⁴⁸ or Gkatzelis et al.²² In cases where the measured species was a subset of the CMAQ-CRACMM species, emissions were not adjusted. Species were generally not adjusted if the inferred model ER was within 30% of the inferred observation ER. In the sensitivity simulation, EF_10,000 was scaled up to a total of 350 g/kg CO, 245 g/kg CO of which could partition between gas and particle phases. The value of EF_10,000 and its uncertainty will be further discussed in the Results. Emission changes were implemented using the Detailed Emissions Scaling, Isolation, and Diagnostic (DESID) module⁴⁹ in CMAQ and summarized in Table S7–S8.

Assessing Potential Public Health Implications

Cancer and noncancer risk from long-term exposure to several ROC_G species in smoke was calculated using methods similar to EPA’s AirToxScreen.⁵⁰ Cancer risk, expressed as additional cancer cases for a population of one million people, or equivalently an N-in-a million risk, was determined using an estimated annual exposure level and unit risk estimate (URE) of toxicity which provides an estimated probability of cancer due to inhalation of 1 μg m⁻³ of the pollutant over a lifetime (Equation S1–S2). URE values for 6 ROC_G species (Table S9) were obtained from EPA.⁵¹ Noncancer (mutagenicity, developmental toxicity, neurotoxicity, and/or reproductive toxicity) risk was estimated by calculating a hazard quotient (Equation S3) for 16 ROC_G species (Table S10) using EPA reference concentration (RfC) values,⁵¹ which are estimates of the concentration that could be inhaled over a lifetime without an appreciable risk. Hazard quotients were summed across mixture species to result in a Hazard Index (HI) (Equation S4).^{50, 52} In general, an HI below 1 indicates air toxics are unlikely to cause adverse health effects.⁵⁰ Risk calculated here assumes a lifetime (implicitly 70 years⁵⁰) of one day of wildfire smoke exposure each year, and estimates scale linearly (up to a factor of 365) to represent additional days per year of smoke exposure over a lifetime (Equation S2).

Standard EPA methods used in AirToxScreen do not provide estimates of risk from inhalation of wildfire PM_2.5 or the full suite of ROC_P as URE and RfC values are not available for these species. In addition, both the measurements and model predictions of ROC_P provide limited composition information. For example, while polycyclic aromatic hydrocarbons (PAHs) are a component of ROC_P, they were not explicitly measured or modeled in this work. While individual components of ROC_P may have differential toxicity, the EPA Integrated Science Assessment for PM determined total mass, rather than any individual PM component or source, is most robustly related with adverse health outcomes.⁵³ In addition, other studies have indicated a potential for woodsmoke to be similar to other PM_2.5 in terms of potential health effects.^{9 54 55} As an initial estimate of cancer toxicity, the median URE value of all polycyclic aromatic hydrocarbons (PAHs),⁵¹ 4.8×10⁻⁵ μg⁻¹ m³, was used as a potential toxicity for all ROC_P. For noncancer endpoints, the EPA RfC for diesel PM (5 μg m⁻³) was used for wildfire ROC_P. In addition, a full range of potential cancer risk was calculated using the most toxic PAH (URE = 1.14×10⁻¹ μg⁻¹ m³) for ROC_P as an upper bound and an estimate of only known PAHs emitted and their toxicity (see Supporting Information) as a lower bound. Similarly, an upper bound estimate for noncancer risk was calculated using the toxicity of benzo[a]pyrene (RfC = 0.002 μg m⁻³) for all ROC_P, and a lower bound considering only the risk of benzo[a]pyrene.

Results and Discussion

Emissions

ROC Gases

Overall, the CMAQ-CRACMM sensitivity simulation with adjusted emissions resulted in an inferred ER (mean ± σ) of 184 ± 9 ppb ROC_G/ppm CO (531 ± 26 g ROC_G/kg CO). This value represents an increase of 71% over the value inferred from the base model with 108 ± 7 ppb/ppm CO (269 ± 20 g/kg CO), which was within 2% of the value across all fires in the 2019 EPA emission inventory indicating consistency in model inputs and the base model inferred ER. The inferred ER in the base model of individual ROC_G species also matched the 2019 EPA inventory⁴² within ~20% on average and within 50% for all species except four reactive alkenes (monoterpenes, isoprene, 1,3-butadiene, and internal olefins) and three oxygenated species (Figure S4). Some discrepancy between the simulation and inventory occurs as the inventory includes fires in different locations with different speciation (Table S2), and model data used to create the inferred ER is sampled coarsely in time (nearest hour) and space (12 km × 12 km horizontal resolution grid box). Major ROC_G species emitted by fires include formaldehyde (13% of observed ROC_G by mole), ethene (10% by mole), acetaldehyde (9% by mole), methanol (8% by mole), terminal olefins like propene (6% by mole), and formic acid (6% by mole).

Emissions of several gas-phase species appeared underestimated in the base CMAQ-CRACMM simulation compared to FIREX-AQ observations (Figure 1). Implementing the cresol (CSL) emission ratio from the work of Gkatzelis et al.²² brought the inferred cresol ER within 20% of the value from FIREX-AQ in this work. In the case of glyoxal + glycolaldehyde (GLY) and methylglyoxal (MGLY) species, the model, even with emissions adjustments intended to result in observed values (Table S7), was unable to produce inferred ER within a factor of 2 of observations and did not show any net production of either species downwind (Figure S3). This result suggests challenges separating emissions and simulating secondary GLY and MGLY in ambient data from fires. Previous work^{48, 56} also suggests glyoxal + glycolaldehyde emissions inferred from ambient FIREX-AQ data are lower than laboratory-based estimates. Sixteen additional species, including the HAPs acetaldehyde (−59% normalized bias in ER for base model), acrolein (−83%), benzene (−66%), 1,3-butadiene (−65%), and toluene (−78%), were low in the base model but remedied in the sensitivity simulation with emission input adjustments. Two species (Very long-lived ROC with k_OH < 3.5 ×10⁻¹³ cm³ molec⁻¹ s⁻¹, which has significant background contributions, and nitroaromatics, which do not have a CRACMM analog) remain low in the simulations and their emissions were not adjusted (Table S7).

Figure 1.

Emission ratios (ER) of ROC_G species relative to CO in ppb per ppm inferred from FIREX-AQ observations and CMAQ-CRACMM predictions. The base (o) and sensitivity (×) simulation values are connected with a vertical line as a visual aid. The grey solid line indicates 1:1 agreement, the dotted lines factor of 2 agreement, and the dashed lines factor of 10 agreement. Red indicates select HAPs. Species are defined in terms of measured and modeled species in Table S5, and values and their uncertainty are available in Table S11. Select species are labeled.

Ten species’ ERs remain higher in the CMAQ-CRACMM sensitivity simulation than observations indicate, and these high values occur when the model species represents a collection of structures that are broader than what was measured. Quickly reacting hydrocarbons (Fast kOH HCs, HC10) in CMAQ-CRACMM are species not otherwise classified with k_OH > 6.8×10⁻¹² cm³ molec⁻¹ s⁻¹ and represented with the chemistry of decane (C₁₀H₂₂). CRACMM workflows place ROC_G emissions unidentified at the individual species level (unspeciated) in this category as a conservatively low estimate of secondary OA and ozone formation potential.³⁹ CMAQ-CRACMM indicates an inferred ER of 18.4 ppb/ppm CO for Fast kOH HCs accounting for about 10% by mole of the overall ROC_G. Semivolatile and intermediate volatility organic compounds not specifically identified (Other S/IROC) are also larger in CMAQ-CRACMM predictions than observations due to poor observational coverage. The increase from the base to sensitivity simulation values in Other S/IROC largely reflects increases in lower volatility emissions that accompany ROC_P (see next section). Other, largely biogenic ROC (Other BROC), <2% by mole of predicted ROC_G, includes monoterpene-, isoprene-, and sesquiterpene-derived aldehydes, peroxides, and nitrates. Other volatile ROC (Other VROC), <4% by mole of ROC_G, includes all other aldehydes, ketones, dicarbonyls, peroxides, and other miscellaneous species for which no measured analog was used. Monoterpene emissions in the CMAQ-CRACMM sensitivity were scaled to target an ER of 5 ppb/ppm CO for all monoterpenes¹⁸ and resulted in an inferred monoterpene ER of 6.5 ppb/ppm CO (<4% of total ROC_G). Note the FIREX-AQ observationally inferred ER from this work of 0.7 ppb/ppm CO is consistent with other observation-based FIREX-AQ estimates²² but is significantly lower than some laboratory-based estimates (2.7 ppb/ppm CO for average of ponderosa and lodgepole pine from FIRELab⁴⁸). The emission process for monoterpenes in fires (distillation) is fundamentally different from that of combustion products⁵⁷ which could lead to some discrepancy.

The ROC_G ER inferred from the CMAQ-CRACMM sensitivity simulation (184 ± 9 ppb ROC_G/ppm CO, 531 ± 26 g ROC_G/kg CO) is approximately double the FIREX-AQ observation-based inferred value (145 ± 69 ppb/ppm CO, 262 ± 13 g/kg CO) in mass units and 27% higher in molar units. The difference in model-predicted vs. observed ROC_G ERs is mainly driven by emissions that are unidentified at the species level (Fast kOH HCs, Other S/IROC) and secondarily to species with known structures that were not measured (many oxidized species, Other VROC, and Other BROC) or thought to be underestimated by the ambient FIREX-AQ observations (monoterpenes). Unidentified emissions tend to be lower volatility and higher molecular weight than traditional VOC emissions and thus contribute relatively more by mass than mole to totals (Table S12, Figure S5). Both FIREX-AQ estimates in Figure 2 (this work and the work of Gkatzelis et al.²²) represent a sum of species measured by mass spectroscopy and chromatography techniques (Table S5) and were not specifically designed to achieve carbon closure. Thus, they represent a subset of all gas-phase emissions. In molar units, the FIREX-AQ ROC_G emissions inferred from high CO conditions in this work are within 7.4% of the value from Gkatzelis et al.²² who used the ratio of maleic anhydride to furan to identify fresh smoke.

Figure 2.

ROC emissions in the gas (ROC_G, blue) and particle (ROC_P, orange) phases in g per kg of CO from this work and other relevant studies by Wiedinmyer et al.²¹ and Gkatzelis et al.²² Error bars represent uncertainty in ROC_P and ROC_G components and span one standard deviation in each direction. Values inferred from ambient data in this work are marked with an asterisk. FINN v2.5 emissions of organic carbon from temperate forests²¹ were converted to ROC_P using an organic matter to organic carbon ratio of 1.85. Values are available in Table S13.

Particulate ROC

Particulate forms of ROC are a substantial contributor to total ROC emissions (Figure 2). The base model simulation with an ER of 54.4 ± 10.5 g of ROC_P per kg of CO emitted closely reflects the 2019 U.S. inventory but is slightly lower due to some evaporation of intermediate volatility species. FINN v2.5 values for temperate forests indicate slightly higher primary organic aerosol emissions of 115 g per kg of CO emitted.²¹ All other ROC_P estimates, which are specific to FIREX-AQ conditions, indicate substantially higher ROC_P emissions, and except for the base simulation, they range from 253 ± 22 g/kg CO in the sensitivity simulation to 449 ± 61 g/kg CO in the observationally inferred estimates of this work.

Since dilution and evaporation are major drivers of ROC_P abundance during FIREX-AQ,³⁰ the model estimates of ROC_P emissions were informed by both the near-source (defined as high CO) as well as downwind organic aerosol concentrations (Figure 3a, downwind gas-phase and total ROC will be discussed in the next section). The base CMAQ-CRACMM simulation indicated underestimates in ROC_P at all levels of CO in smoke. In addition, the base model showed a constant or slightly increasing ROC_P to CO ratio as CO decreased, indicating net OA formation. This result contrasts with the observed trend which indicates ~450 g ROC_P/kg CO in fresh smoke which steadily decreases towards a value of ~100 g/kg CO and indicates evaporation causes decreases in ROC_P that are stronger than dilution alone. Secondary OA production does not appear to be a driver of near-source (<10 hours and up to a day³⁰) changes in ROC_P abundance for this study in contrast to other work that has shown substantial SOA production from long-time aging.⁵⁸

Figure 3.

Evolution of (a) particle-phase ROC in mass units, (b) gas-phase ROC in molar units, (c) gas-phase ROC in mass units, and (d) total ROC in mass units for wildfire smoke. All quantities are relative to the background (Δ). Observations (Obs) are from FIREX-AQ measurements and estimated as a sum of species for ROC_G when all measured species are available. Inferred emissions ratios relative to CO (ER) (points with error bars, numbers in ppb/ppm CO (molar units) or g/kg CO (mass units)) are shown at the 99th percentile in CO. Error bars for the ER show ±1 standard deviation (y-axis) with the values for ROC_G propagated from individual species. Error bars for the ER on the x-axis indicate the range of CO from 97th percentile to maximum in smoke. Shading on the lines represents the interquartile range. Model values are from a CMAQ-CRACMM Base and sensitivity (Sens) simulation.

Emissions of species with C* ≤ 10⁴ μg m⁻³ (species covered by EF_10,000) were increased by 4.8× in the sensitivity simulation to reasonably match ROC_P at all CO levels probed by the model (Figure 3a) but resulted in a different estimate of ROC_P emissions than the observationally inferred value for two reasons. First, the CMAQ-CRACMM model does not predict levels of CO as high as the near-source observations (Figure S3) indicating the model may experience excessive dilution and/or transport compared to observations. Some amount of excessive dilution is expected in the model as fire emissions are distributed throughout a 12 km × 12 km grid box at the time of emission. Second, absorptive partitioning is a non-linear process so dilution will non-linearly affect the abundance of the condensed organic phase. Increasing EF_10,000 to 350 g/kg CO in the sensitivity simulation brought the observations and model into closer agreement than in the base case and corrected the model trend such that it shows ROC_P decreases downwind beyond what is expected from dilution alone.

Some discrepancies remain in the downwind trend of ROC_P in the sensitivity simulation compared to observations. The sensitivity simulation does not show as strong a decrease in ROC_P/CO with decreasing CO as the observations and even shows relatively constant ROC_P/CO near source. The CMAQ-CRACMM sensitivity predicts all species with the ability to condense (species with C* ≤ 1,000 μg m⁻³) are condensed in the very fresh smoke. If the C* ≥ 10,000 μg m⁻³ classes of species were allowed to partition, CMAQ-CRACMM could produce a stronger decrease of ROC_P/CO with decreasing CO and a ROC_P ER more consistent with observations (albeit at a lower CO as the model does not experience high CO). The steepness of the ROC_P/CO vs. CO data suggests the model should partition C* 10,000 μg m⁻³ compounds in fresh smoke to better capture near-source trends. However, increasing the total EF_10,000 above the sensitivity simulation is not well supported given the limitations in capturing concentrated CO conditions in the model. The FIREX-AQ observations are relatively consistent with an EF_10,000 of ~500 g/kg CO in the work of Pagonis et al.,³⁰ but that value resulted in model overestimates compared to observations at almost all CO levels despite slightly underestimating the total ROC_P emissions (Figure S6).

Total ROC

The CMAQ-CRACMM sensitivity simulation results in an inferred ER for total ROC of 784 ± 34 g/kg CO which is consistent with the 711 ± 62 g/kg CO inferred from FIREX-AQ observations here which are slightly higher than previous FIREX-AQ work^{22, 59} (Figure 2). Recall that the gas-phase FIREX-AQ measurements were targeted towards specific compounds so do not include any species unidentified at a compound (or at least formula) level and are often a subset of some compound classes. In addition, ROC_P emissions in the CMAQ-CRACMM sensitivity remain low compared to observations, and 105 g/kg CO of the EF_10,000 emissions input to the model are unable to condense. If that potentially misallocated 105 g/kg CO (which is part of ROC_G in Figure 2) was relabeled as particle phase, the model ROC_G would still be about 60% higher than the FIREX-AQ observationally inferred ROC_G ER. FINNv2.5 temperate forest gas-phase emissions (with unidentified species) support the CMAQ-CRACMM sensitivity simulation ROC_G estimates as they match within 16% by mass or 72 g/kg CO. Thus, the CMAQ-CRACMM sensitivity is considered the most complete estimate of ROC_G and total ROC while still allowing that some amount of mass around a C* of 10,000 μg m⁻³, on the order of about 100 g/kg CO, is inadequately constrained and/or potentially missing from the model.

Figure 2 emphasizes that gas and particle ROC are two components of ROC. The exact balance of ROC_P vs. ROC_G in total emissions will be determined by temperature, degree of dilution, and other organic aerosol sources in the environment where the emissions occur. Partitioning between phases cannot explain the difference between total ROC emissions in standard EPA inventories and observations. The 2019 inventory and base CMAQ-CRACMM simulation indicate 320 g/kgCO and 324 ± 22 g/kgCO of total ROC emitted, lower than observations of organic aerosol alone from FIREX-AQ. Overall, standard EPA inventories underestimate ROC emissions from fires in this work by about a factor of 2× with ROC_P emissions being low by a factor of 5–8× depending on whether the final CMAQ-CRACMM sensitivity value or inferred FIREX-AQ value is compared to the base model.

To facilitate preparation of emission inventories, two subsets of ROC defined at a threshold in C* of 320 μg m⁻³ have been proposed: condensible (CROC) for the lower volatility and gaseous (GROC) for the higher volatility group.⁶⁰ This definition avoids operationally determining the phase of emissions at specific conditions which may or may not be known and may differ from those where the ambient release occurs. In the CMAQ-CRACMM sensitivity simulation, a CROC emission factor of 210 g/kg CO was imposed. This value is a factor of ~3 higher than what the 2019 EPA inventory reports for primary OA and a factor of ~2 higher than FINNv2.5 organic particulate emissions. From the sensitivity simulation, GROC is then 574 g/kgCO, about 2× the base gas-phase ROC inventory and higher than all ROC_G values in Figure 2 (Figure S7) due to the inclusion of intermediate volatility compounds which may be condensed at the point of emission.

Emissions of ROC from fires are a significant source of carbon. CMAQ-CRACMM ROC_P emissions are slightly less oxygenated than FIREX-AQ observations indicate with an organic matter to organic carbon ratio of 1.52 in the sensitivity simulation (1.60 in base simulation) vs 1.74 in observations. This indicates marginally better agreement in carbon units for the inferred ROC_P ER in the sensitivity simulation (389 ± 40 gC/kgC in CO) vs observations (612 ± 97 g C/ kg C in CO) than when mass units are used. In terms of total ROC, the CMAQ-CRACMM sensitivity predicts an ER of 1,250 ± 60 gC/kgC in CO and the observations indicate 978 ± 94 gC/kgC in CO. Thus, both the observations and model predictions indicate at least as much carbon, if not 25% more, is emitted as ROC as is emitted as CO from fires. The CMAQ-CRACMM base simulation reflects roughly half of that emitted carbon mass (480 ± 37 g C/kg C in CO) (Table S14). Considering fires emit ~2 Pg C yr⁻¹ and CO is ~10% of that value,^{20, 61} this suggests global fires may emit ~200 Tg C yr⁻¹ (~300 Tg yr⁻¹) of ROC, similar to some emission inventories^{18, 62} but about a factor of 4 higher than previous estimates from a global model.⁶³ Lower ROC emissions, such as those in the base simulation and some field campaigns,⁶⁴ imply a lower estimate of ~100 Tg C yr⁻¹.

Downwind ROC

FIREX-AQ data best capture smoke within multiple hours of emission (less than a photochemical day; see analysis of Liao et al.²⁵). Both model simulations as well as FIREX-AQ observations indicate roughly a factor of 2 increase in moles of ROC_G relative to CO, from near-source (high CO) to downwind (low CO) conditions (Figure 3b). Both the observations and sensitivity simulation show increases on the order of 20 to 40 ppb/ppm CO each for methanol, formaldehyde, and acetone which largely explain the overall downwind increase (Figure S8). All three species show clear enhancements relative to background at the time of emission (Figure S3), and they are also common fragmentation products from a variety of precursors. Formaldehyde has a short lifetime against photolysis, and secondary formaldehyde quickly dominates over primary formaldehyde from fires within a few hours of atmospheric aging.²⁶ A previous analysis of FIREX-AQ data indicated substantial formaldehyde production with an effective yield of 0.33 formaldehyde molecules produced per molecule of precursor oxidized.²⁵ Other fragmentation products are enhanced downwind relative to CO but less consistently. The observations show a >40 ppb/ppm CO increase in formic acid which is not present in the model. Ethanol (model and observations), acetic acid (model and observations), methylglyoxal (observations), aldehydes (model and observations), PANs (model and to a lesser degree observations), and MEK (model) approach about a 10 ppb/ppm CO increase downwind (Figure S3, S8). Other VROC, which includes functionalized compounds with no measured equivalent, is enhanced downwind by 80 ppb/ppm CO in the model sensitivity simulation.

While molar units emphasize the formation of fragmentation products downwind consistent with previous laboratory experiments,^{65, 66} mass-based ROC_G units indicate either steady or decreasing abundance relative to CO (Figure 3c). The observations and base model indicate roughly steady ROC_G relative to CO downwind while the sensitivity simulation indicates decay relative to dilution. ROC_G species that show large downwind decreases for the sensitivity simulation also contribute substantially to total mass emissions: Other S/IROC, Fast kOH HCs, and Monoterpenes (Figure S8). Combined, these species decrease by 174 g/kg CO downwind. Some of this mass is retained in fragmentation products, and Other VROC increases by 89 g/kg CO downwind. Note that trends become more sensitive to background corrections as CO decreases (see Figure S3), and above 40 μg m⁻³ of CO the sensitivity simulation indicates roughly constant dilution-corrected ROC_G. Carbon-based units show similar trends as the mass-based ROC_G units (Figure S9).

Total ROC mass across gas and particle phases decreases downwind relative to CO in the observations and sensitivity simulation (Figure 3d), with the model-predicted trends in ΔROC/ΔCO with ΔCO more moderate in their decrease than observations. The observations likely overestimate the true decrease in ΔROC/ΔCO with decreasing CO as the evaporated OA is not measured once it is in the gas phase. Specifically, ~300 μg/kg CO of the observed downwind decrease in ROC_P does not appear to be added to the observed ROC_G downwind. The sensitivity simulation tracks a larger amount of total ROC downwind but still indicates loss compared to emissions. While CRACMM was designed to improve how total emitted mass was captured in chemical transport models, the chemistry does not yet conserve mass as some products (such as carbon dioxide) are not tracked and the use of model surrogates of a specific carbon number necessitates approximations in mass.³⁹ For example, Fast kOH HCs, which decrease by nearly 70 g/kg CO downwind, have 10 carbons in their representative structure, but their organic nitrate and peroxide products are represented by structures with 4 and 2 carbons, respectively.

Toxicity and Risk

Chronic (lifetime) exposure to species within ROC can lead to cancer and noncancer health effects, and ambient exposure to wildfire smoke has been linked with increased rates of cancer in Canadian⁶⁷ and European⁶⁸ studies. Figure 4 shows the estimated health risk of chronic exposure to the gas and particle ROC from wildfire smoke based on observations during FIREX-AQ with estimates on the primary y-axis for 1 day of annual exposure each year over a lifetime. Both cancer and noncancer risk on a per capita basis are highest for fresh emissions and predicted to decrease in downwind smoke due to dilution. Note that over a given region, many more people may be exposed to the downwind concentrations versus fresh conditions. For the initial ROC_P toxicity estimate, total ROC risk for both cancer and noncancer endpoints is dominated by the particle phase, and risk decreases downwind slightly stronger than dilution alone would indicate due to evaporation of ROC_P (see Figure S10 for dilution-corrected trends).

Figure 4.

Estimated risk of ROC gases (blue) and particles (orange) based on observed concentrations during FIREX-AQ for (a) cancer and (b) noncancer endpoints. All quantities are relative to the background (Δ). Cancer risk (left y-axis) is expressed as the increase in cancer cases per million people for each day per year of lifetime exposure to fire smoke. Noncancer risk (left y-axis) is expressed as the increase in hazard index (HI) for each day per year of lifetime exposure to fire smoke. Points (× symbols, values labeled in primary y-axis units) with error bars are based on freshly emitted smoke and shown at the 99th percentile in CO. Error bars indicate the range of CO from 97th percentile to maximum in smoke (x-axis). Darker shading represents uncertainty in risk for one standard deviation variability in species concentrations. Light shading for ROC_P risk represents an upper- and lower-bound range resulting from toxicity assumptions (see Methods and Supporting Information). The secondary y-axes (right) indicate (a) the number of days per year of lifetime smoke exposure required to reach a one-in-10,000 risk of cancer and (b) the number of days per year of lifetime smoke exposure to reach an HI of one with the horizontal grey dashed line at 100 days of exposure for reference.

For cancer, the initial assumption of using the median PAH toxicity value for all ROC_P results in the particle phase driving 98% of the estimated risk for freshly emitted smoke with 200 additional cancer cases for each day of annual lifetime exposure in a population of one million people compared to 3 cases per million per day of annual lifetime exposure for gas-phase ROC. Gas-phase ROC contributors to cancer risk at the time of emission (Figure S11) include formaldehyde (61% of ROC_G emission cancer risk), benzene (10%), acetaldehyde (10%), 1,3-butadiene (9%), acrylonitrile (5%), and naphthalene (4%) in agreement with previous work¹³ indicating formaldehyde is the dominant driver of cancer risk for gas-phase species in wildfire smoke. Even though the estimate of downwind ROC_G cancer risk declines relative to emissions (Figure 4a), the dilution-corrected cancer risk of ROC_G slightly increases compared to emissions (Figure S6a) due to downwind production of secondary formaldehyde. Despite these downwind changes, the particle phase continues to dominate downwind ROC cancer risk (>95%). Even if wildfire ROC_P were 10× less toxic than initially assumed (URE of 4.8×10⁻⁶ μg⁻¹ m³), it would still drive more than 80% of the cancer risk of all ROC in fresh emissions and 66% of the downwind risk. At a factor of 100× lower toxicity than initially assumed for ROC_P, which puts risk near the lower bound estimate of ROC_P risk based only on PAHs with known toxicity (light orange), the gas phase would become the dominant contributor of ROC cancer risk, but ROC_P would still meaningfully contribute.

The 1989 Benzene National Emission Standard for Hazardous Air Pollutants (NESHAP) Rule established 1-in-10,000 (equivalently 100 cases per million people) as the upper limit of acceptable ambient air inhalation cancer risk for the most exposed population.⁶⁹ The secondary y-axis in Figure 4a estimates the number of days of annual, lifetime smoke exposure to reach the 100-in-a million cancer risk. When smoke is relatively fresh and concentrated (>2,450 ppb CO), one day of annual exposure over a lifetime to smoke ROC_P is predicted to result in a risk above 1 in 10,000 for the initial ROC_P URE. At 100 days of annual lifetime exposure (grey dashed line Figure 4), a long season for a firefighter⁷⁰ and a duration of exposure at some U.S. locations,^{71, 72} downwind ROC_P levels examined here could exceed the 1 in 10,000 level of risk. ROC_G gas-phase HAP levels at low CO (Figure 4a) generally do not exceed the 1 in 10,000 risk threshold. However, 100 days of exposure to fresh ROC_G emissions is estimated to contain enough HAP mass to result in a 300-in-a million cancer risk.

For noncancer endpoints, both gas-phase HAPs and ROC_P are estimated to contribute competitively to risk (Figure 4b). Noncancer risk in Figure 4b is expressed as the increase in hazard index (HI) for each day of annual smoke exposure (primary axis), and the secondary axis indicates the number of days annual exposure to reach a HI of one, a level at which the potential for adverse noncancer health effects increases. ROC_G contributors to noncancer risk (Figure S12) include acrolein (72% of ROC_G emission noncancer risk), hydrogen cyanide and acetonitrile (10%), acetaldehyde (7%), formaldehyde (7%), 1,3-butadiene (2%), and other species including methanol and aromatic compounds (<2%) consistent with previous work.¹³ In fresh emissions at high CO, ROC_P contributes 74% of the overall ROC noncancer risk. For a factor of 10× lower toxicity than initially assumed (RfC of 50 μg m⁻³, similar to acetonitrile, Table S10), ROC_P would contribute 22% of the noncancer risk of emissions. Given the lower bound estimate of ROC_P noncancer risk, factors of 100× lower effective toxicity are not plausible. One day of exposure to fresh ROC emissions just reaches an estimated HI of one (considering uncertainty) with downwind exposures being lower and resulting in HI values less than 1. Continuous, daily exposure (365 days per year over a lifetime) to low levels of fire smoke (<35 ppb CO), is not estimated to result in an HI above 1 unless ROC_P is much more toxic than the initial assumption.

Considerations for Future Work

This work suggests a large, but uncertain, potential for long-term exposure to ROC from wildfires to lead to cancer risk. Given the potential for ROC_P to contribute to health risk from fires, better constraining the toxicity of ROC_P in future work would improve our understanding of wildfire risks to health. The tendency of ROC_P to evaporate raises questions about the role of PM components in exposure and health impacts. The persistent and growing gap between total ROC and ROC_P as plumes dilute suggest that toxicity and risk is lost as ROC_P evaporates. Whether this reflects a true differential health impact of a compound depending on its phase when it enters the body, or whether this simply results from gaps in understanding of the toxicity of individual compounds remains an important area for future research. Improved methods for determining the role of ROC species in adverse health outcomes would lead to a better understanding of the role of wildland fires in long-term human health and inform strategies to effectively mitigate adverse impacts of ROC and other compounds. In addition, FIREX-AQ emissions of ROC_P tend to be higher than some previous work³⁰ which highlights the need for continued characterization of ambient ROC_P levels and understanding drivers of variability.

Synopsis:

Understanding the long-term health effects of chronic exposure to wildland fire smoke will require improved information on the composition and toxicity of reactive organic emissions and their downwind changes.