Main

Volatile chemical products (VCPs), including cleaning agents, personal care products, coatings, pesticides, printing inks and adhesives, contain complex mixtures of organic species that span a wide range of volatilities and oxidation states. The evaporation of components, such as organic solvents, results in emissions of volatile organic compounds (VOCs), primarily in indoor environments1. Direct exposure to these VOCs can result in adverse health effects, including mucosal irritation, endocrine disruption, asthma and non-specific symptoms such as headache2,3,4,5,6. Furthermore, these VOCs escape into ambient air, where they contribute to the formation of criteria pollutants such as fine particulate matter (PM2.5) and ozone (O3)1, imposing further risks to public health7,8,9. Because of the dramatic reduction in tailpipe emissions in the United States and Europe, recent estimates suggest that VCPs may be the largest source of urban VOC emissions1. However, gaps remain in our understanding of the impacts of VCPs on air quality and human health.

Organic aerosol (OA) is a substantial component of ambient PM2.5 (ref. 10), with a large fraction (typically >60%) formed in the atmosphere from secondary chemistry11,12. Regional chemical transport models struggle to simulate PM2.5, specifically the organic component formed by photochemical activity (that is, secondary OA (SOA)) in urban areas13,14,15,16,17,18. Li et al.19 report high PM2.5 yields for some individual intermediate volatility compounds found in VCPs and consumer product mixtures. Siloxanes, widely used in consumer products and commonly observed in urban air20, can have PM2.5 yields of 20–50% (refs. 21,22). Underestimated VCP emissions, including species that are semivolatile to intermediately volatile1,23 or highly oxygenated24, are probably responsible for some of the missing organic PM2.5 in regional air quality models.

Given the potential contribution of VCPs to both near-field (nominally indoor, during or in the vicinity of product use25) and far-field (ambient air) exposure pathways, an approach that integrates their resulting impacts is needed. While most direct exposure to VCPs for people26 will be in the indoor environment, ecosystem and other environmental exposure (for example, secondary pollutants) resulting from VCPs will be dominated by the far-field environment27,28,29. Far-field exposure due to residential and commercial releases of VCPs and their resulting criteria pollutant formation are not considered in current chemical exposure estimates26.

In this work, we show that a model developed for quantifying near-field chemical exposure due to consumer product use can be used to estimate ambient air emissions for far-field air quality modelling. Implementing higher ambient chemical emission estimates in a regional-scale chemical transport model results in organic PM2.5 and ozone predictions that are more consistent with ambient measurements. These simulations demonstrate that VCPs are important precursors to criteria pollutants and that ambient air measurements can be used to constrain near-field chemical exposure models.

Results

Estimating the impacts of VCP use on air quality and public health requires information on (1) organic product usage (the amount of product used that consists of carbon-containing compounds), (2) emission to air (described by a population-normalized emission factor, \(\overline {{\mathrm{EF}}_i}\); see Methods) and (3) transformation pathways leading to secondary products. Existing near-field chemical exposure models (for example, the High-Throughput Stochastic Human Exposure and Dose Simulation, SHEDS-HT, hereafter called SHEDS (ref. 30)) help constrain the first two items, while far-field ambient chemical transport models (for example, the Community Multiscale Air Quality (CMAQ) modelling system) require the latter two as input. Thus, ambient air quality modelling, when evaluated against observed air concentrations, can improve the data, assumptions and methods on which near-field chemical exposure models rely.

Organic product usage and emissions to air

A first step in constraining the contribution of VCPs to air pollution is better quantification of their air emissions to the environment. Figure 1a compares different ambient air emission estimates (filled bars, see also Supplementary Tables 1 and 2 and Methods). In the United States, the evaporation of chemicals from organic products is estimated in the Solvent Sector as part of the US Environmental Protection Agency (EPA) National Emissions Inventory (NEI) and EPA Solvent Tool. Figure 1a indicates that these operational NEI methods result in a factor of ~3–4 lower emissions of VCP organic gases per person than a posteriori estimates (Methods) derived from the near-field SHEDS model30 and values presented by McDonald et al.1.

Fig. 1: Comparison of organic product usage, EFs and PM2.5 yields across different methodologies.
figure 1

a, Organic product usage per person \(\overline {S_i}\) (total height of the bars including unfilled portion) and \(\overline {{\mathrm{EF}}_i}\) values attenuated by washing down the drain (filled portion of the bars). The \(\overline {S_i}\) and \(\overline {{\mathrm{EF}}_i}\) values labelled ‘CA’ are California-specific; the other values are nationwide estimates. \(\overline {S_i}\) for household products in SHEDS (one unfilled bar) and the total \(\overline {S_i}\) and \(\overline {{\mathrm{EF}}_i}\) (the rightmost bar cluster) values are ×5 higher than the y axis as indicated by the ×5 label; the asterisks indicate that data are not available. Also note that \(\overline {S_i}\) estimates are not shown for the California component of the 2011 NEI (only \(\overline {{\mathrm{EF}}_i}\) values are available for the California component). b, PM2.5 yields by mass organic gases reacted estimated by McDonald et al.1 and upper bound estimates for prompt formation in the CMAQ modelling system. The PM2.5 yield of 5% by mass of organic gases was adopted in the CMAQ feasible solution; however, a PM2.5 yield of 10% (represented by the star on the far right) also resulted in PM2.5 SOA concentrations consistent with measurements

Source data.

Full size image

Even though SHEDS and McDonald et al. use very different input data and approaches, the a posteriori SHEDS model and McDonald et al. arrive at approximately the same nationwide average air EFs of ~20 and ~24 kg per person per yr, respectively. The SHEDS-estimated values of \(\overline {{\mathrm{EF}}_i}\) for personal care are consistent with those of McDonald et al. (that is, 5.2 versus 4.5 kg per person per yr), and both methods estimate consumer adhesive emissions at ~2 kg per person per yr. However, SHEDS has the highest \(\overline {{\mathrm{EF}}_i}\) for household products, resulting in roughly similar amounts of organic gas emissions as personal care, while McDonald et al. indicates that personal care emissions are the highest among VCP subsectors (2× higher than household cleaning). The \(\overline {{\mathrm{EF}}_i}\) values for California are generally lower than the nationwide ones in both the 2011 NEI (4.7 versus 8.2 kg per person per yr) and McDonald et al. (10.8 versus 24.3 kg per person per yr) estimates, potentially reflecting differences in down-the-drain adjustments applicable to certain categories31 and more stringent regulations for coatings and adhesives in California32,33. Summing the per capita EFs for all sectors available in SHEDS and multiplying by population results in estimates of total residential (excluding industrial and commercial) consumer product emissions of organic gases to air of 1.4 Tg yr−1 in the 2011 NEI, compared with 6.1 Tg yr−1 (~4×) in SHEDS and 3.9 Tg yr−1 (~3×) in McDonald et al. for the United States. In addition to these numbers that suggest the NEI is low on a nationwide basis, we will show that an increase in the California component of the NEI to effectively 14 kg per person per yr (CMAQ feasible solution; the purple filled bar on the right in Fig. 1a) also results in improved estimates of criteria pollutants and other air quality endpoints in California, as predicted by a regional-scale chemical transport model.

Differences in organic product usage drive the differences in high versus low emission estimates. The Solvent Tool and EPA NEI methodology assume complete volatilization and air release of organic components in products (\(\overline {S_i}\) = \(\overline {{\mathrm{EF}}_i}\)) yet have lower \(\overline {{\mathrm{EF}}_i}\) values than other methods that assume only a fraction of the organic product evaporates (Fig. 1a). This means that lower organic product usage (the total height of the bars in Fig. 1a) in the EPA national-level methodology results in lower \(\overline {{\mathrm{EF}}_i}\) values than in SHEDS and McDonald et al. These discrepancies in organic product usage are particularly large for sectors such as personal care, household products and consumer adhesives, with SHEDS and McDonald et al. indicating ~10–20× higher organic product usage than the NEI methods. Total organic product usage for the VCP sector is estimated at ~75, ~42 and ~8 kg per person per yr in the SHEDS, McDonald et al. and 2011 NEI methodologies, respectively. These differences indicate that constraining the organic product usage is of high priority for creating robust emission estimates.

The SHEDS and McDonald et al. approaches result in high air emissions, and volatilization assumptions modulate the magnitude of emissions. In this work, organic species in SHEDS were volatilized into air on the basis of species vapour pressure and nature of application, resulting in an effective volatilization fraction (γ, see Methods) of 27% in a posteriori estimates (Supplementary Table 3). For household products, a large fraction of the organic species (~86% in McDonald et al. and ~89% in SHEDS) were predicted to go down the drain, consistent with the types of product applications in that category, such as laundry detergent and cleaning products. SHEDS (a posteriori) indicates that half of the organic compounds in personal care products are ultimately emitted to air. This volatilization fraction is lower for SHEDS (γ = 0.48) than for McDonald et al. (γ = 0.73), reflecting different assumptions about the types of applications and/or composition of personal care products. In most other categories, the down-the-drain adjustment is less than a factor of 2 (that is, 0.5 < γ < 1). Looking across all VCP categories available, ~58% and ~27% of the mass evaporated, according to the McDonald et al. and SHEDS methodologies, respectively.

A posteriori SHEDS estimates show an increased importance of inhalation exposure versus dermal uptake and/or ingestion from consumer product usage compared with a priori estimates before the modifications described in the Methods. The value of γ was increased from 2% in the a priori SHEDS chemical exposure model to 27% a posteriori, by making more reasonable assumptions for chemical mass retained on surfaces and skin (Methods). These results illustrate that volatilization assumptions, particularly for dermally applied personal care products, can substantially affect air release estimates. The 10× increase in γ between a priori and a posteriori implies a 10× increase in post-use inhalation exposure. Using Isaacs et al.30 as a baseline and factoring in the a posteriori volatilization, the inhalation pathway is now estimated to be responsible for 2% of the mean daily intake dose for volatile chemicals in the near-field environment, making it competitive with previous estimates of dermal (7% of dose) and object-to-mouth (3% of dose) but not hand-to-mouth (near 90% of dose) exposure30. This estimate of the role of inhalation in near-field chemical exposure applies to consumers in residential buildings, and the relative importance of inhalation exposure could be higher for populations in other environments (for example, commercial or industrial settings). Volatilization should be further explored to more consistently represent the importance of chemicals in both chemical exposure and ambient air quality applications.

Speciation and PM2.5 yields in a regional air quality model

Since VOCs can be precursors for secondary PM2.5, another step in determining the role of VCPs in air pollution is to identify transformation pathways leading to secondary products. The predicted ambient air PM2.5 yield depends on the composition of the emissions and the yields of individual species. Individual species in regional chemical transport models such as EPA’s CMAQ modelling system are represented using simplified chemical mechanisms (Methods), which typically use a reduced number of surrogates to represent VOCs. Existing simplified chemical mechanisms were not developed to provide a robust representation of VCP chemistry (Fig. 2).

Fig. 2: Organic gas emissions and PM2.5 SOA formation potentials for personal care products based on current and new emission composition, respectively, using parameters in CMAQ.
figure 2

The blue shaded portion indicates that the current operational method was used to specify emission composition, while updated information was used for the yellow panel. Emission values (in g) that do not correspond to the left y axis are displayed in black text. PM2.5 yields by mass of organic gases are indicated with red figures.

Source data

Full size image

The challenges associated with representing PM2.5 are illustrated in Fig. 1b, which shows the prompt (rapidly formed, within several hours) sector-specific PM2.5 yields based on a current model system that employs these surrogate methods (see also Supplementary Tables 4–6). The PM2.5 yields for VCP subsectors in CMAQ vary by an order of magnitude (Supplementary Table 7), ranging from 0.17% for personal care products to as much as 16% for some industrial coatings, with an overall effective value of about 3.5% by mass in California (Methods). The CMAQ overall PM2.5 yield for VCPs is slightly lower than the estimate from McDonald et al. (that is, 4.6%). However, if the VCP sector is required to fully close the CMAQ model-measurement gap in PM2.5 SOA, then an even higher yield of 10% is required (discussed in the next section).

If the differences between the CMAQ and McDonald et al. (Fig. 1b) PM2.5 yield estimates are considered, in combination with the previously discussed emission magnitude discrepancies between SHEDS/McDonald et al. and the NEI (Fig. 1a), then adhesive, pesticide and personal care product uses account for the biggest potential underestimations in nationwide predictions of PM2.5 mass formation from VCPs at 75 Gg yr−1, 40 Gg yr−1 and 35 Gg yr−1, respectively (Supplementary Table 7). This total missing PM2.5 from these three subsectors (140 Gg yr−1) is comparable to SOA formation from mobile sources (~170 Gg yr−1) and accounts for 50% of VCP SOA (~300 Gg yr−1)1. The potential underestimate of organic PM2.5 from pesticides and adhesives is driven almost entirely by the emissions magnitude, while the aerosol yield (15× higher) causes the difference for personal care products. As a result, personal care is probably the category where PM2.5 underestimates are driven largely by an incorrect characterization of VOC emission composition across the entire ambient air quality modelling system.

Figure 2 illustrates how the PM2.5 yield from personal care emissions is underestimated in CMAQ and how it can be improved. A small fraction of the emitted organic gases (~1.7%) are mapped to traditional PM2.5 SOA precursors including alkanes, aromatics and monoterpenes. Default emission processing and mapping results in an effective PM2.5 mass yield of 0.17 g per 100 g of organic gases. SOA from personal care products in CMAQ is formed mainly from alkane oxidation, which contributes ~57% of the PM2.5, followed by aromatic oxidation (~38%).

Resolving issues associated with the chemical composition of consumer product emissions, especially compounds with lower volatility that are poorly represented in the current CMAQ mechanism model surrogates34, can increase the PM2.5 yield for personal care products by a factor of 13 (from 0.17% to 2.3% by mass of organic gases, the fifth bar in Fig. 2). Updating VOC speciation profiles on the basis of more recent surveys of consumer products from 1997 to 2010 in California35 results in a ~3× increase in amount of traditional SOA precursors.

The biggest cause of the underestimated PM2.5 SOA formation from personal care VCPs is the lack of surrogate species representing low-vapour-pressure species with high PM2.5 formation potential. For instance, n-heptadecane, a compound present in personal care products with a PM2.5 yield of >30% (ref. 19), is mapped to an alkane surrogate that has low SOA formation (~3% by mass in CMAQ). Additionally, substantial amounts of emitted organic gases are mapped to lumped species labelled as “non-reactive organic gases” and “nonvolatile” species (probably semivolatile or intermediate-volatility organic compounds considered ineffective ozone precursors) that are not considered in CMAQ regional model calculations of ozone or PM2.5. Accounting for contributions of “non-reactive organic gases”, such as siloxanes22,36, and treating the “nonvolatile” species as precursors, such as n-pentadecane with a ~50% (refs. 37,38) (lower bound) or 100% PM2.5 yield (upper bound, Fig. 2), increases the PM2.5 yield estimated for personal care products by a factor of 4–5 (from 0.44% to as much as 2.3%), suggesting that chemical mechanisms need to be extended to include these non-traditional species.

Constraining emissions and PM2.5 yields with ambient observations

Because the SHEDS and McDonald et al. inventories suggest that the 2011 NEI underrepresents VCPs, emission adjustments and PM2.5 yields were implemented in CMAQ to investigate what combinations of parameters improved predictions of ozone, PM2.5 and other metrics for the Los Angeles basin. CMAQ-simulated ambient PM2.5 SOA concentrations in Pasadena, California, are driven by both the VCP emission magnitude and the PM2.5 yield, as reflected by the normalized mean bias (NMB) between the measured and predicted SOA (Fig. 3). The current estimate of VCP emissions in the 2011 NEI (California) and CMAQ model (that is, VCP emissions of 4.7 kg per person per yr + effective PM2.5 yield of ~3.5%) results in an underestimation of photochemically driven prompt SOA of 53% (SOAprompt = SOA with background contributions removed; see Methods). Using the existing SOA parameterization and simply increasing VCP emissions across all subsectors by factors of 2–4 increases SOA production and improves model performance. We consider the 14 kg per person per yr of VCP emissions (3× NEI (California) emissions; 1.7× higher than the 2011 NEI nationwide total \(\overline {{\mathrm{EF}}_i}\) estimate) and a 5% PM2.5 yield as a plausible far-field modelling representation of the VCP sector (feasible solution). Yields could be higher or lower in combination with lower or higher emissions, and the feasible solution still underestimates SOA by 22%. With 2011 NEI emissions, some bias remains even with a PM2.5 yield of 10% by mass from organic gases. Additional SOAprompt sources as well as background sources, probably of biogenic or non-fossil origin, may still be missing in CMAQ (Extended Data Fig. 1).

Fig. 3: NMB between CMAQ-predicted and observed SOAprompt in Pasadena, California, for different combinations of VCP emission magnitude and VCP PM2.5 yield adjustments.
figure 3

The cross hatching indicates that the parameters are inconsistent with observations (specifically, NMB ≥ uncertainty in measured OAs, that is, 38% (ref. 67)). The recommended emissions (top) and PM2.5 yields (right) in McDonald el al. are marked by the range indicated in the margin. The recommended emission magnitude based on SHEDS (consumer emissions only) is indicated with the red asterisk (top). Two cases, the current model (dashed) and feasible solution (solid), are indicated with yellow boxes. The base VCP emissions in the California component of the 2011 NEI are 4.7 kg per person per yr.

Source data

Full size image

The feasible solution case, supported by the near-field SHEDS model emission estimates and updated VOC speciation profiles, demonstrates that VCPs probably play an important role in PM2.5 SOA formation. The 14 kg per person per yr VCP emissions with a PM2.5 yield of 5% per mass organic gas also substantially improve the predicted diurnal variation of prompt SOA with a peak concentration of 4.7 μg m−3 (versus 2.4 μg m−3 in the operational CMAQ modelling system) at Pasadena (Fig. 4). However, the simulated SOA was still lower than the observed SOAprompt, which reached a diurnally averaged maximum of 6.6 μg m−3. The base CMAQ system estimated that VCPs contributed 0.2 μg m−3 of PM2.5 SOA, compared with 1.0 μg m−3 from non-VCPs (Supplementary Table 8), which means that VCPs are not a major source of SOAprompt (only ~20%) in the current models. In the feasible solution case, VCPs are predicted to contribute 1.1 ± 0.3 μg m−3 of PM2.5, making up 53% of the simulated total SOAprompt (2.1 μg m−3) and ~41% of the observed SOAprompt at Pasadena (2.6 μg m−3). The strong daytime increase in SOA requires higher PM2.5 yields regardless of the emission magnitude (Extended Data Fig. 2).

Fig. 4: Diurnal variation of simulated and observed SOAprompt at Pasadena, California.
figure 4

The simulation with VCP emissions zeroed out is indicated by the dashed line. The grey boxes indicate the 25th and 75th percentiles of the observations, with the median values marked with horizontal lines in the boxes, and the whiskers extend to the 10th and 90th percentiles. The solid lines indicate the means of the observations and simulations.

Source data

Full size image

VCP use results in large amounts of organic gases being emitted into ambient air and subsequently affects oxidant levels in addition to PM2.5. In the feasible solution case, VCPs account for ~3.7 ppb of organic gases in ambient air during the daytime (10:00–18:00). Increased organic gas emissions from VCPs trigger modest increases in hydroxyl radical (OH) and O3 concentrations in the model (Extended Data Fig. 3 and Fig. 5a), which results in faster hydrocarbon reactions and more SOA. Increased VCP emissions increase the high bias in simulated OH concentrations (NMB of 29% versus 21% in the current model), but still within the uncertainty of the measurement (that is, 36%)39. Organic gas emissions from VCPs also contribute 9 ± 2 ppb to the maximum daily 8 hr average (MDA8) O3 with the feasible solution (versus 3 ppb in the current model; Fig. 5a). The higher VCP emissions improve the model simulations of O3 compared with observations, resulting in the feasible solution regression line overlapping the 1:1 line at Pasadena.

Fig. 5: Simulated MDA8 O3 and OH reactivity versus observations at Pasadena.
figure 5

a,b, MDA8 O3 (a) and OH reactivity (b) with different VCP emission magnitudes. In a, the simulation with VCP emissions zeroed out is indicated with pink, and the averaged MDA8 O3 concentrations are given in the parentheses.

Source data

Full size image

OH reactivity is a measure of how quickly OH is lost via reaction with organic gases and is thus useful for assessing uncertainties in VOC emissions and examining their importance to O3 and PM2.5 formation40,41,42,43. OH reactivity was enhanced in the feasible solution case (15.5 s−1 versus 12.8 s−1 in the current model) and was brought to within 30% of the observation (21 s−1, ref. 39) with 3× the NEI (California) VCP emissions (Fig. 5b). OH reactivity contributed by VCPs (3.9 s−1 in the feasible solution) is higher than predicted in the default CMAQ modelling system but lower than 4.8 s−1 in McDonald et al. The OH reactivity of 4.5–4.7 s−1 is explained by organic gases from non-VCP sources (such as mobile sources and biogenics). Additionally, inorganic species account for ~6.9 s−1 of OH reactivity, higher than the observation-based value (5.8 s−1) due to overestimated NO2 concentrations in the CMAQ simulations. Overall OH reactivity has better agreement with measurements with higher VCP emissions, but some missing OH reactivity in CMAQ remains to be addressed.

Integrating near-field and far-field exposure models

Near-field chemical exposure (such as SHEDS) and far-field air quality (such as CMAQ) models share common goals in terms of providing estimates of human (and ecosystem) exposure to pollutants. When connected through the CMAQ model, ambient air concentrations of OA, ozone and OH reactivity provide top-down constraints on air emissions associated with VCP use and suggest an EF of ~14 kg per person per yr in California (feasible solution). This value is consistent with a posteriori SHEDS calculations of air emissions of ~20 kg per person per yr. SHEDS indicates that this ambient air emission rate is associated with ~75 kg per person per yr of organic product usage, notably higher than the current EPA national-level methodology, which indicates only 8.2 kg per person per yr. In addition to using ambient air concentrations to constrain emissions and organic product usage in near-field modelling, our work suggests additional avenues to better estimate exposure across scales. For example, the EPA has undertaken a major effort to curate hundreds of thousands of chemical structures via the CompTox Chemistry Dashboard44; however, over 250 species in the ambient air emission speciation database used here (Methods) were not found in the Dashboard (as of 9 September 2019), although chemical identifier curation is ongoing. The curation of species emitted in air in the Dashboard will allow for quantitative structure–activity relationship models developed for chemical properties45, toxicity46 and exposure26 to be applied and facilitate risk evaluations of these compounds. In addition, models like SHEDS can be used to create new speciation profiles for ambient air quality modelling or provide alternative EFs for use in creating VOC inventories. Ultimately, connecting near-field chemical exposure and far-field air quality models that consider direct exposure to emitted chemicals as well as their oxidation products (for example, PM2.5 and ozone) will result in improved tools for addressing public health questions.

Discussion

VCPs are ubiquitous in daily life for aesthetic, protective, health-care and other purposes. Because VCPs have the potential to directly expose people to harmful chemicals as well as escape to ambient air, their formulations have continuously evolved over decades in response to air quality and chemical safety drivers. Since the 1990s, VCPs have been targeted for control to reduce the formation of ozone as well as hazardous air pollutants in the United States via the Clean Air Act. This has led to an increase in the prevalence of water and other green solvents in VCPs. Product formulations will continue to evolve to improve performance, cost and regulatory compliance via the use of functional substitutes as chemicals with lower health risks and less potential for criteria pollutant formation are identified. However, designing product reformulations aimed at improving public health and the environment will require a wholistic perspective that accounts for multiple pathways of exposure to both the compounds themselves and their degradation products, which is not fully realized in current alternative assessment tools or air quality management. For example, a narrow focus on ozone has led to compounds such as siloxanes being exempt from VOC regulation despite their high potential for SOA formation. Multiple options, from solvent substitution to product capture and recycling, should be employed to reduce the environmental and health impacts of VCPs.

Methods

In this work, air emissions of VCPs were estimated with multiple inventory methods and a near-field chemical exposure model. The impacts of VCP emissions on ambient criteria pollutants were evaluated via air quality modelling, taking uncertainties in emissions and the model representation of pollutant formation into account (Extended Data Fig. 4).

VCP emission estimates for ambient air

The use of VCPs results in evaporative emissions of organic gases to the atmosphere. Air emissions from VCP usage (E) can be estimated on the basis of sector-specific EFs (EFi) and activity data (Ai) for sector i:

$$E = \mathop {\sum}\nolimits_i {{\mathrm{EF}}_i \times A_i}$$
(1)

Emissions can be estimated using different activity surrogates such as employment, population or product use in a sector with appropriate EFs (Supplementary Table 2). If population-based EFs are reported by the method (such as Solvent Tool and SHEDS), they are directly assigned to \(\overline {{\mathrm{EF}}_i}\) (Fig. 1a). If population-based EFs are not provided, the values of \(\overline {{\mathrm{EF}}_i}\) (in units of kg per person per yr) are computed using E divided by Ai in the form of population (using 312 million as the 2011 US population and 3.8 million for California). These \(\overline {{\mathrm{EF}}_i}\) values quantify how many kilograms of organic gas are released to ambient air per person each year. EFs may consider how organic product usage (Si) (mass per person per year for a population-based activity surrogate, that is, \(\overline {S_i}\)) is attenuated by washing down the drain (assumed here to be a permanent sink of VOCs) or otherwise sequestered from air. This down-the-drain adjustment is represented by γi, which represents the fraction of the organic product (generally a solvent) released into ambient air (0 ≤ γi ≤ 1):

$${\mathrm{EF}}_i = S_i \times \gamma _i$$
(2)

The work here is focused exclusively on the organic content of consumer products that can be released into the atmosphere. The water, electrolytes and other inorganic components of consumer products are not explicitly discussed or included in Si, γi or EFi. In this work, five different methods (see also Supplementary Table 2) for calculating VCP EFs are compared in terms of total sector emissions and category-specific emissions. These methods are described in the following sections.

The 2011 EPA NEI

The EPA NEI is compiled on the basis of air emission estimates developed by state and local agencies and submitted to the EPA, as required by the Air Emissions Reporting Requirements Rule 40 CFR § 51.1008 (2019). In the cases where state or local agencies do not provide emission estimates, the EPA provides them as a fallback47. The California VCP emission estimates in the NEI are contributed by the state agency using its own methodology (‘The California emission inventory’). The 2011 EPA NEI estimates are national-level VOC emission totals aggregated up from state-level reports (including not only non-point-source but also point-source emissions, which are available only for industrial sectors such as industrial surface coating, graphic arts and degreasing, and may use methods from individual jurisdictions or EPA methodology). The EPA NEI Solvent Sector emissions are used synonymously with VCP emissions in this work.

The fallback 2011 EPA NEI methodology is based on national-level estimates of solvent use from the Freedonia 2007 report48 divided by national-level activity data such as population and employment, as well as some other sources (for example, the EPA Emissions Inventory Improvement Program technical report49). The default EPA NEI methodology assumes complete volatilization of the organic components (γ = 1). The NEI inventories VOCs according to a strict definition based on what is likely to lead to ozone formation (40 CFR 51.100). Conversion between the inventoried VOCs and total organic gases (which may include non-reactive or other species) is based on other information (for example, SPECIATE; see below). The difference in regulatory VOCs versus total organic gases is small (<30% effect) compared with the differences between the 2011 NEI/Solvent Tool and other methods, which are generally on the order of a factor of 3 (Supplementary Table 9 and Fig. 1a).

The Solvent Tool

The Solvent Tool v.1.7 uses the same methods as the 2011 NEI and has been distributed by the EPA since the development of the 2014 NEI to provide a way for states to estimate emissions of VOCs from VCPs. EFs in the Solvent Tool v.1.7 for some categories were updated from 2011 using the updated Freedonia 2013 report50. The differences in EFs between the 2011 NEI and the Solvent Tool are negligible (Fig. 1a). EFs in the Solvent Tool apply to non-point emissions only50.

The California emission inventory

The California Air Resources Board (CARB) estimates both total organic gas and VOC emissions from VCPs using the sales of products from consumer product surveys from product manufacturers and formulators, multiplied by the weight fraction of each organic species (VOC) compound (that is, the VOC content of the products) according to the speciation data from the survey31. The aggregated category (such as hair sprays or hand and body lotion) total organic gases (VOCs) are computed as a sum over each individual product organic gases (VOCs) total, followed by adjustment with values of γ in certain categories (0 < γ < 1).

The SHEDS model

The SHEDS model v.0.1.6 (ref. 30) and its default inputs were used to estimate the residential release of VOCs from formulated consumer products. Ambient air quality (for example, CMAQ) models generally target the actual atmospheric mean state, while exposure models (for example, SHEDS) are designed to inform the risk-based prioritization of chemicals via conservative (upper-bound) exposure estimates for individual species26,30. SHEDS uses consumer habits and practices (product prevalence in the marketplace, frequency of use and mass per use) for 324 individual product categories, and predicts the uses of these products for a population of simulated individuals representative of the United States in terms of age and gender (a standard population size of 10,000 was used here). Releases of individual chemicals were estimated by multiplying the mass of the product used by the median weight fractions from Material Safety Data Sheets obtained from the EPA’s Chemical and Products Database51 and by values of γ (see below). Mean releases were estimated for all volatile chemicals for which weight fraction data were available. Organic gas emissions for general categories (Supplementary Table 10) were calculated by summing over all chemicals in the category. All volatilized product was assumed to be released into ambient air without considering air exchange rates or loss in the indoor compartment.

We use the terminology a priori to refer to SHEDS predictions using the standard configuration described by Isaacs et al.30. Two critical adjustments were made to standard SHEDS operations (that is, a priori estimates), resulting in the a posteriori estimates shown in Fig. 1a: (1) the product prevalence in the marketplace was adjusted to be more mass conservative across all VOCs and (2) γ was estimated as a function of species vapour pressure and type of product application. By default, SHEDS estimates exposure to individual compounds, and if a species is present in one product, the model will always choose a non-zero value for that species’ occurrence in that product type (100% chemical prevalence). For example, if only one shampoo contains siloxanes, that product is assumed to have 100% of the market share when siloxane exposure is being estimated. Similarly, if a different shampoo is the only one that contains polyethylene glycol, that shampoo will have 100% of the market share when polyethylene glycol exposure is being estimated. This assumption is made due to the lack of true knowledge of the fraction of products on the market containing the chemical and the market share of those products. However, it can result in the overcounting of mass when individual species are aggregated into a VOC or organic gas total. To remedy this, we opted to randomly sample product compositions, even across products in the Chemical and Products Database in which a species was not present (zero). The species were then aggregated up to a total. Furthermore, while SHEDS applies down-the-drain assumptions for products applied to residential and body surfaces, chemicals applied dermally do not volatilize by default. Volatilization was parameterized on the basis of the type of usage (for example, applied to surfaces versus used in an aqueous environment) and the volatility of organic species (following the methodology of McDonald et al., Supplementary Table 11) to obtain emissions to ambient air.

Recent estimates in literature

McDonald et al.1 estimated total organic gas emissions from VCPs using the annual sales of products from economic and commodity flow surveys, the organic content usage in each product based on energy and chemical production statistics, and the volatilization fraction of the organic content based on indoor air quality literature and emission control efficiency.

VCP emission processing in air quality models

The composition of VCP emissions (referred to as speciation) in air quality modelling is based on sector-specific speciation profiles from the EPA’s SPECIATE database52. This database provides information to convert inventoried VOCs to specific individual chemical species. SPECIATE v.4.5 is the public version of the database in operational use at the time of this work, and consumer product composition is based mainly on the CARB 1997 consumer products survey. For SPECIATE v.5.0 (ref. 35) (released in summer 2019), the EPA updated the profiles for consumer products and architectural coating using CARB’s 2010 consumer products inventory products update, which is based mainly on the 2003 and 2006 CARB consumer product surveys. Current and new emission compositions in Fig. 2 are based on SPECIATE v.4.5 and SPECIATE v.5.0, respectively. The speciation profiles for emission sources such as consumer products and architectural coatings, derived from CARB profiles, include a number of species (for example, bins) that are not defined in SPECIATE. Those were mapped to compounds or mixtures in SPECIATE v.5.0 using Carter’s assignments53 to avoid loss of mass. In addition, we updated the exempt status of VOCs relevant to VCPs in SPECIATE v.5.0 so that the ratio of total organic gases to inventoried non-exempt VOCs is better characterized. Mappings of chemical composites to model-ready surrogates, which are chemical-mechanism-dependent, follow Carter34,53.

CMAQ model configurations and inputs

The CMAQ model is an ambient air chemical transport model used for research and regulatory purposes. For example, it is used by state and local agencies to design State Implementation Plans and other air quality management strategies. In this work, the CMAQ model uses VCP emissions as input to predict their impact on ambient concentrations of ozone, SOA and OH reactivity. CMAQ model v.5.3Beta (www.epa.gov/cmaq, https://github.com/USEPA/CMAQ) was used to perform simulations in California (Supplementary Fig. 1) with a horizontal resolution of 4 km and 35 vertical layers extending up to the 50 mbar level in the atmosphere. The simulation spanned the California Nexus of Air Quality and Climate Change (CalNex) campaign54 (15 May to 15 June 2010) with the prior 14 days as a spin-up. The Pasadena ground site was located 18 km northeast of downtown Los Angeles and impacted by emissions in the Los Angeles metropolitan area with prevailing southwesterly winds during the daytime (advection of 3–5 h)15. The SAPRC07tic chemical mechanism (see the details in Pye et al.55 and Xie et al.56) was used to represent gas-phase chemistry, with AERO7 for aerosol chemistry57,58,59. The meteorological inputs were provided by Weather Research and Forecasting v.3.8.1, with model configurations identical to those in Lu et al.60

The anthropogenic emissions used in CMAQ were based on the 2011 NEI, with a 25% reduction of mobile NOx to achieve more consistency with the observed NO2 at the Air Quality System sites within the domain (Supplementary Fig. 2). The biogenic emissions were calculated online using the Biogenic Emission Inventory model v.3.61, with the Biogenic Emission Landuse Database 4 and multiple data sources for vegetation representation including the US Forest Service Forest Inventory and Analysis v.5.1 for 2002–2013 and the US Department of Agriculture 2007 census of agriculture data61.

Treatment of primary OA and SOA from non-VCPs in CMAQ

Direct emissions of organic PM2.5 (primary OA) for area-wide anthropogenic emission sectors (gasoline vehicles, diesel vehicles, aircraft, cooking, biomass burning, dust, VCPs and other) were characterized as semivolatile with sector-specific volatility profiles for the mobile60 and non-mobile14 sources. Atmospheric processing of primary OA is the same as described in Murphy et al.57. A state-of-the-science PM2.5 SOA parameterization for mobile organic gases with intermediate volatility was implemented in CMAQ v.5.3, as in Lu et al.60, in an effort to minimize uncertainty in model predictions of the mobile-sector SOA. Other sources of SOA in the base model include the oxidation of traditional VOCs including isoprene, monoterpenes, sesquiterpenes, aromatics and alkanes. For computational efficiency, SOA from VOCs of primarily anthropogenic origin (benzene, toluene, xylene, polycyclic aromatic hydrocarbons and alkanes) were parameterized in CMAQ v.5.3-AERO7 as described in Supplementary Tables 4 and 5.

Treatment of SOA from VCPs in CMAQ

In addition to a base CMAQ v.5.3 simulation, a series of simulations were performed (Supplementary Table 12) with VCP organic gas emission perturbations (1×, 2×, 3× and 4× relative to the base VCP emissions in the 2011 NEI). Three levels of fixed PM2.5 yields (α in Reaction 1) including 1%, 5% and 10% by mass were explored in concert with the emission perturbations. Due to the diversity of VOCs emitted and created in the atmosphere (Goldstein and Galbally62 estimated that 104–105 different species have been identified), lumped mechanisms (for example, SAPRC07) are created to represent the major routes to ozone and PM2.5 formation with a reduced number of organic species (usually ~100 surrogates). SOA yields for VCP subsectors in Figs. 1b and 2 were calculated offline (that is, not conducting CMAQ modelling) by mapping the emission composition to SAPR07 surrogates with the surrogate-specific yields in CMAQ (Supplementary Table 6). Simulated SOAprompt with base CMAQ v.5.3 was equivalent to a simulation with a fixed PM2.5 yield of ~3.5%. Organic gases from VCPs participated in gas-phase chemistry leading to ozone formation and oxidant modifications through traditional SAPRC mechanism species (that is, benzene, toluene, xylene, monoterpenes, etc.) as indicated by SPECIATE 4.5, while the SOA formation from VCPs was imposed by Reaction 1. In Reaction 1, all VCP organic gases were treated using a single surrogate, TOGVCP, so that a specific yield (α = 1%, 5% and 10%) could be implemented. The reaction rate constant (kOH) for VCP organic gases reacting with OH in Reaction 1 is identical to the averaged kOH for mobile organic gases with intermediate volatility in CMAQ (ref. 60)—that is, kOH = 2.18 × 10−11 cm3 per molecule per second, which is slightly higher than the McDonald et al. estimate of 1.25 × 10−11 cm3 per molecule per second (ref. 1).

$${\mathrm{TOG}}_{{\mathrm{VCP}}} + {\mathrm{OH}} \to \alpha {\mathrm{SOA}} + {\mathrm{OH}}$$
(Reaction 1)

CMAQ ambient air quality model evaluation

The CMAQ and Weather Research and Forecasting models were extensively evaluated for the CalNex domain for multiple pollutants (for example, ozone, total PM2.5 mass and constituents, and PM2.5 precursor vapours) in previous work14,16,57,60,63,64,65. The evaluation in this study was designed to focus on rapidly formed photochemically driven SOA (that is, SOAprompt), to isolate fresh SOA from VCPs in an urban atmosphere. Oxygenated OA (OOA), derived from factor analysis of OA data from the Aerodyne High-Resolution Time-of-Flight Aerosol Mass Spectrometer, is considered a measured proxy for SOA and can be further mathematically divided into two components: semivolatile OOA (SV-OOA) and low-volatility OOA (LV-OOA)12. Generally, SV-OOA shows strong diurnal variation (peaking at 14:00–15:00) in Pasadena and is consistent with fresh SOA from urban areas, while LV-OOA (the more oxidized OOA) appears flat throughout the day and is expected to be aged SOA (ref. 66). The observed background OA was derived from the minimum in the LV-OOA diurnal circle15 and is probably associated with biogenic VOC oxidation and transport. Here, the average of the lower 50% of CMAQ-simulated SOA concentrations during 0:00–4:00 local time was used to estimate the background OA in simulations. SOAprompt, in both the model predictions and the observations, was derived by subtracting the background OA from the overall SOA concentration. The background adjustment could lead to negative values of prompt OA on an hourly basis, which has minor effects on diurnal cycles or period-averaged concentrations (Supplementary Fig. 3). A metric NMB was used for the evaluation of simulated SOAprompt with CMAQ:

$${\mathrm{NMB}} = \frac{{\mathop {\sum}\nolimits_{i = 1}^N {(M_i - O_i)} }}{{\mathop {\sum }\nolimits_{i = 1}^N O_i}}$$
(3)

where Mi and Oi are the modelled and observed values at time i, respectively, and N is the number of hourly samples.

OH reactivity from VCP emissions in the simulations was estimated using the concentration of the surrogate TOGVCP with single kOH (Reaction 1). The values calculated with this method compared well with the values calculated using SAPRC07 speciation (Supplementary Fig. 4). The measured total OH reactivity was based on the observed decay of OH due to reactivity with ambient air in a turbulent-flow tube reactor.39