The Recent and Future Health Burden of Air Pollution Apportioned

Mar 18, 2013 - Using photochemical source apportionment modeling and a health ...... Expanding Health Technology Assessments to Include Effects on the...
0 downloads 0 Views 3MB Size
Download PDF
Policy Analysis pubs.acs.org/est

The Recent and Future Health Burden of Air Pollution Apportioned Across U.S. Sectors Neal Fann,*,1 Charles M. Fulcher,1 and Kirk Baker1 1

Office of Air Quality Planning and Standards, U.S. Environmental Protection Agency, 109 T.W. Alexander Drive Research Triangle Park, North Carolina 27711, United States S Supporting Information *

ABSTRACT: Recent risk assessments have characterized the overall burden of recent PM2.5 and ozone levels on public health, but generally not the variability of these impacts over time or by sector. Using photochemical source apportionment modeling and a health impact function, we attribute PM2.5 and ozone air quality levels, population exposure and health burden to 23 industrial point, area, mobile and international emission sectors in the Continental U.S. in 2005 and 2016. Our modeled policy scenarios account for a suite of emission control requirements affecting many of these sectors. Between these two years, the number of PM2.5 and ozonerelated deaths attributable to power plants and mobile sources falls from about 68 000 (90% confidence interval from 48 000 to 87 000) to about 36 000 (90% confidence intervals from 26 000 to 47 000). Area source mortality risk grows slightly between 2005 and 2016, due largely to population growth. Uncertainties relating to the timing and magnitude of the emission reductions may affect the size of these estimates. The detailed sector-level estimates of the size and distribution of mortality and morbidity risk suggest that the air pollution mortality burden has fallen over time but that many sectors continue to pose a substantial risk to human health.

■

photochemical model.12 This approach allowed us to attribute ambient annual mean PM2.5 and summer season (May to September) ozone levels at a 12 km grid resolution across the Continental United States to each of 23 categories of emission sectors in 2005 and 2016; these sectors account for all inventoried anthropogenic and biogenic emissions affecting PM2.5 and ozone levels in the Continental U.S. The 23 sectors include nine industrial point sources (cement kilns, coke ovens, electric arc furnaces, ferroalloy facilities, iron and steel facilities, integrated iron and steel facilities, pulp and paper facilities, refineries, and all other point source emissions); electricity generating units (EGU); four area sources (residential wood combustion, taconite mining, re-entrained fugitive emissions from industrial sources and all other area sources); four mobile sources (on-road, nonroad, aircraft/locomotives/marine vessels, and ocean-going vessels); two international sources (Canada/Mexico and trans-boundary emissions); wildfires, secondary organic aerosols, and biogenics. We stratified the sectors in this way to ensure that we both represented the spatial heterogeneity in air quality impacts among sectors, and

INTRODUCTION The epidemiological, toxicological, and clinical literature have convincingly demonstrated the relationship between ambient tropospheric ozone, fine particles, and the risk of adverse health outcomes.1−3 The epidemiological literature in particular has extensively characterized the relationship between populationlevel exposure to these two ubiquitous pollutants and the risk of an array of acute and chronic adverse health impacts, including premature death, hospital visits for respiratory and cardiovascular illnesses, and the exacerbation of asthma, among many others.4−7 An increasing number of quantitative risk assessments have applied risk coefficients reported in this literature to estimate the overall burden to public health attributable to recent levels of PM2.5 (Particulate matter, 2.5 μm or less in diameter) and ozone globally and in the Continental U.S.8−11 Such analyses give valuable insight to the size and spatial distribution of air pollution impacts on human health due to recent air quality levels, but they tell us less about the emission sectors contributing most to this burden or how this contribution could change over time. We build upon the literature above by attempting to attribute the total PM2.5 and ozone air quality and health burden to a host of stationary, area, and mobile sectors located in the U.S. and globally. We ran the source apportionment module in the Comprehensive Air Quality Model with Extensions (CAMx) This article not subject to U.S. Copyright. Published 2013 by the American Chemical Society

Received: Revised: Accepted: Published: 3580

November 27, 2012 March 13, 2013 March 18, 2013 March 18, 2013 dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

also to reflect the expected reduction in PM2.5 and ozone levels due to regulatory requirements affecting certain sectors (EGUs and mobile sources in particular). Using the environmental benefits mapping and analysis program (BenMAP), which contains a library of concentration−response relationships and demographic data necessary to perform a health-impact assessment, we quantified the number of premature deaths, chronic and acute illnesses due to PM2.5 and ozone from these 23 sectors for each analytical year.13 For these sectors we also assessed the geographic and age distribution of health impacts, characterizing the change in this distribution between 2005 and 2016, assuming the implementation of several national air quality rules during this time period.

■

MATERIALS AND METHODS We quantified the number of PM2.5 and ozone-related deaths and illnesses using a health impact function, which incorporates information regarding air quality levels, population exposure, baseline health data and a risk coefficient. The approach we follow to calculating health impacts assessment has been applied extensively in the literature and is methodologically consistent with several recent quantitative risk assessments.11,14−16 As an example of this calculation, a log−linear health impact function estimating changes in incidence as a result of changes in air quality would look as follows:

Figure 1. Conceptual diagram of steps in health impact calculation for 2016.

Δy = y0 (e β·Δx − 1) ·Pop

provided by the U.S. EPA Science Advisory Board 20we also consider the potential influence of this assumption. Emissions and Air Quality Modeling. Photochemical Air Quality Modeling. Our first step was to quantify PM2.5 and ozone levels attributable to each of 23 sectors. We applied a state of the science 3D Eulerian photochemical transport model to simulate the complex processes related to the formation, transport, and fate of ozone and PM2.5. CAMx includes treatment of emissions, wet and dry deposition of gases and particles, gas-phase chemistry, aqueous-phase chemistry, and particulate matter formation through inorganic and organic pathways.12,21−24 The CAMx model has been used extensively for regulatory modeling of both pollutants and has received extensive peer review.12,25−27 We applied CAMx version 5.30 using a full year of 2005 meteorology and an emissions inventory that accounted for both 2005 and 2016 emission levels; below we describe how we projected emission levels to 2016. We selected the year of 2005 because it matched an available national emissions inventory year and because the meteorology for that year was conducive (e.g., summer stagnant air masses with clear skies) to ozone and PM2.5 formation. We performed photochemical modeling using domains that cover the United States with 12 km square sized grid cells and that stratify the vertical atmosphere with 14 layers, most of which are nearest the surface, to resolve the diurnal variation in the boundary layer; the height of the lowest model layer is approximately 30 m. The 36 km domain covers a larger spatial extent than both of the 12 km domains and is used to downscale air quality predictions from an annual 2005 simulation of the Goddard Earth Observing System Chemistry (GEOS-Chem) model.28 The use of a global model to provide boundary inflow allows for more realistic representation of PM2.5 and ozone transported from outside the model domains. Additional information related to photochemical model options, meteorological modeling used as input to the

Where β is the risk coefficient drawn from an epidemiological study for the health end point of interest, y0 is the baseline incidence rate for that end point, Δx is the change in air quality, and Pop is the population of interest. We applied BenMAP v4.0.52 to automate this process, as it contains the population, concentration−response relationships and baseline health data needed to calculate the health impact function. BenMAP has been applied in several recent risk assessments.17,18 Briefly, in this assessment we characterize PM2.5 and ozone-related premature deaths, hospital admissions, emergency department visits, and cases of acute respiratory symptoms, among other impacts occurring throughout all populated areas in the Continental U.S. for which we modeled PM2.5 and ozone levels. In the interest of space, we place in Supporting Information (SI) full details regarding the chronic and acute health end points we assessed, the effect coefficients and baseline incidence rates we applied and the steps we followed to calculate health impacts and changes in life years; these parameters are consistent with the values reported in a recent U.S. EPA analysis.19 Because air pollution health impact assessments are wellestablished in the literature, we focus here on our novel use of source apportionment modeling to generate the air quality surfaces used in this analysis; the SI describes in detail how we specified the health impact assessment. Figure 1 depicts the sequence of analytical steps we followed to attribute air quality levels and health impacts to each sector. We also acknowledge that there is growing evidence that PM2.5 composition, which varies across regions and by season, may affect its toxicity. However, in our judgment the literature does not yet yield data that could be used as the basis for a risk assessment that quantitatively accounted for such differences.2 Thus, while in this assessment we assume that particles from each source are equally toxican approach that is consistent with recent advice 3581

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

photochemical model, and performance evaluation is provided in the SI. Emissions Modeling. We used the 2005 National Emissions Inventory to estimate anthropogenic emissions.29 We estimated biogenic emissions using hour-specific meteorology as input to the Biogenic Emissions Inventory System (BEIS) model. Mexican emissions (1999) and Canadian emissions (2006) are identical for the 2005 and 2016 simulations, allowing us to isolate the impact of changes in U.S. anthropogenic and biogenic emissions.30,31 Other international emissions are included in the model simulations through boundary inflow and are identical for both 2005 and 2016. We projected 2016 emissions from the 2005 emissions using economic and population growth factors for some sectors and accounted for reductions required by emissions control programs and expected plant shut-downs. We projected electricity generating unit emissions to 2016 using the Integrated Planning Model (IPM), which the EPA has applied in a number of recent regulatory analyses for the EGU sector and is documented elsewhere.32 The point source inventory in 2016 reflects emissions changes related to multiple Federal control programs: the proposed Cross-State Air Pollution Rule (CSAPR) affecting emissions from electricity generating units, the maximum achievable technology (MACT) standards for industrial boilers, the MACT for reciprocating internal combustion engines, and the NOx state implementation call.33−36 The 2016 air quality modeling simulations do not reflect emission reductions anticipated from the EGU sector as a result of the recently promulgated mercury and air toxics standards, and so are likely to overstate the PM2.5 and ozone contribution from this sector and do not account for the vacatur of the CSAPR rule. 37 Attributing PM2.5 and Ozone Concentrations to Sources. Source contribution techniques implemented in photochemical grid models such as CAMx are used to estimate the contribution of emissions sources to model estimates of ambient concentration. Contribution information is similar to observation based source apportionment approaches such as positive matrix factorization or chemical mass balance receptor models. These observation based receptor models are limited to times/places where measurements are made and have difficulty differentiating sources that have similar patterns of emissions. Air quality models do not suffer from these limitations and can be used to provide more specific distinction between source and receptors. Particulate matter precursor emissions are tracked in CAMx to secondarily formed sulfate and nitrate. Sulfur dioxide, sulfuric acid, and primarily emitted PM2.5 sulfate are tracked to PM2.5 sulfate. Nitrogen oxides, nitrous oxides, and primarily emitted PM2.5 nitrate are tracked to PM2.5 nitrate. Primarily emitted organic carbon, elemental carbon, and soil/crustal species are tracked directly to the associated ambient specie. CAMx has several options for ozone source apportionment, and for this analysis we use the anthropogenic precursor culpability assessment (APCA). This algorithm directs ozone contribution to VOC or NOX sources based on the estimated ozone formation regime, unless the biogenic contribution is limiting in which case the model redirects the contribution to the anthropogenic source category.12 Photochemical model source apportionment has been widely used by regulating agencies and peer reviewed in the literature,38−41 and the strengths and limitations of this approach are discussed elsewhere.42

We selected sectors for tracking to provide a broad delineation of the total inventory into biogenic and anthropogenic categories. These sectors represent a diverse set of emissions groups in terms of emissions totals of ozone and PM2.5 precursors and proximity to National Ambient Air Quality Standard (NAAQS) nonattainment areas and population. The 2005 NEI and projected future year do not contain emissions for outdoor wood boilers, suggesting that the contribution estimates for residential wood combustion is likely a lower bound estimate. The secondary organic aerosol category is a total modeled estimate of PM2.5 organic carbon formed through secondary reactions from anthropogenic and biogenic emissions at each receptor and is not attributable to any specific source group. As a result, the contributions from sectors through SOA are understated. The contribution from lateral boundary inflow was based on hourly CAMx output from the 36 km domain that completely covers the continental United States, ensuring that boundary contributions account for all emissions from outside the United States and the portions of Canada and Mexico that are outside the 12 km model domains. Post Processing Contributions. The model estimated ozone and speciated PM2.5 originating from each sector noted below in each 12 km sized grid cell for each hour. Two separate 12 km domains were used and grid cells located in both were averaged before subsequent analysis. We aggregated hourly ozone values to average summer season 8 h maximum and aggregated hourly PM2.5 species to annual averages to be consistent with the effect coefficients used to quantify health impacts. Modeled PM2.5 contribution estimates were expressed as a percentage of bulk modeled speciated PM2.5 (i.e., predictions from all sectors) to estimate annual average relative response factors for each grid cell. Relative response factors for each PM2.5 species were matched by grid cell with FRM monitor locations using the modeled attainment test software (MATS) program.43 The species specific RRFs (relative response factor) representing sector contribution are multiplied to monitored PM2.5 composition from colocated or nearby PM2.5 speciation monitors to generate species specific contribution estimates. The constituent species are summed to generate an estimate of contribution to total PM2.5 design values. The MATS program also makes adjustments to the modeled concentrations to match measurement and analytical artifacts known to exist using federal reference methods (FRM) to measure total PM2.5 mass.44 These manipulations are largely related to changing modeled PM2.5 ammonium nitrate to account for volatilization off Teflon FRM filters during warmer months, assuming sulfate and nitrate ions are nearly fully neutralized with ammonium, and estimating particle bound water using an empirical regression equation based on concentrations of sulfate, nitrate, and ammonium ions. Similarly, ozone source contribution estimates were expressed as a percentage of bulk modeled ozone to estimate daily relative response factors for each grid cell, which were averaged over all modeled elevated air quality days.45 Ozone design values are multiplied by the relative response factor from the grid cell containing the monitor to estimate source contribution at that location. Rather than point estimates of air quality and sector contribution, full model surfaces of annual average PM2.5 and average summer season ozone are input to BenMAP to estimate health impacts. The MATS program creates a fused surface of annual average PM2.5 and summer season ozone using observation data to minimize model 3582

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

Figure 2. Distribution of sector contributions at all monitors to annual PM2.5 (left) and maximum 8 h ozone (right) for current year 2005 and projected year 2016.

Figure 3. Number of premature PM2.5- and ozone-related deaths attributable to seven broad classes of sectors in 2016 (thousands of premature deaths).

and error are bounded by 200%, which is the poorest model performance. All of the statistical metrics presented approach 0 when agreement between observed and model predicted values are best. Observations are matched with model predictions using the grid cell where the monitor is located. Model predictions and observations are paired matching the temporal resolution of the observations before aggregate performance metrics are estimated.46 The modeling systems slightly underestimates annual aggregate total PM2.5 (mean bias = −0.7 ug/m3 and fractional

prediction outliers that may be related to instances where meteorology or emissions are not well characterized.

■

RESULTS CAMx Model Evaluation. Photochemical model output is compared with speciated PM2.5 measurements taken from 402 monitors locations (SI) that are part of two different monitor networks: IMPROVE and CSN. Recommended statistical metrics for performance evaluation include mean bias, mean error, fractional bias, and fractional error.45,46 Fractional bias 3583

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

health outcomes

3584

12 000 (8 800, 15 000) 31 000 (17 000, 45 000) 13 000 (5 000, 22 000) 4 200 (2 100, 7 100) 4 600 (2 800, 6 500) 6 800 (210, 13 000) 9 500 000 (8 100 000, 11 000 000) 420 000 (170 000, 660 000) 1 600 000 (1 400 000, 1 800 000) 710 000 (44 000, 3 600 000) 19 000 (−620, 37 000)

420 (190, 660) 1 900 (1 400, 2 400) 3 900 (1 100, 6 500) 1 400 (0, 4 300) 2 600 000 (1 300 000, 4 000 000) 860 000 (390 000, 1 200 000)

industrial point

31 000 (22 000, 39 000) 79 000 (43 000, 110 000) 33 000 (12 000, 53 000) 10 000 (5 300, 17 000) 11 000 (6 700, 16 000) 18 000 (560, 35 000) 25 000 000 (21 000 000, 29 000 000) 1 100 000 (440 000, 1 700 000) 4 300 000 (3 700 000, 4 800 000) 2 100 000 (120 000, 11 000 000) 480 000 (−1 600, 95 000)

460 (200, 720) 2 100 (1 600, 2 700) 4 200 (1 200, 7 000) 1 600 (0, 4 900) 3 100 000 (1 600 000, 4 600 000) 990 000 (440 000, 1 400 000)

area

17 000 (12 000, 22 000) 44 000 (24 000, 63 000) 19 000 (7 200, 31 000) 6 000 (3 100, 10 000) 6 800 (4 000, 9 500) 9 500 (300, 19 000) 13 000 000 (11 000 000, 15 000 000) 560 000 (230 000, 880 000) 2 200 000 (1 900 000, 2 400 000) 1 100 000 (59 000, 5 800 000) 25 000 (−840, 49 000)

380 (170, 580) 1 7 00 (1 300, 2 200) 3 500 (950, 5 800) 1,200 (0, 3 700) 2 300 000 (1 200 000, 3 400 000) 750 000 (330 000, 1 100 000)

electricity generating units

17 000 (12 000, 21 000) 43 000 (23 000, 62 000) 18 000 (6 700, 29 000) 5 600 (2 900, 9 600) 6 200 (3 700, 8 700) 9 800 (310, 19 000) 14 000 000 (12 00 0 000, 16 000 000) 610 000 (250 000, 970 000) 2 400 000 (2 100 000, 2 700 000) 1 100 000 (65 000, 5 400 000) 27 000 (−920, 55 000)

2 300 (1 000, 3 600) 11 000 (7 800, 13 000) 21 000 (5 700, 34 000) 7,800 (2, 23 000) 15 000 000 (7 500 000, 22 000 000) 4 800 000 (2 100 000, 6 800 000)

mobile

9 900 (7 200, 13 000) 25 000 (14 000, 37 000) 11 000 (3 900, 17 000) 3 300 (1 700, 5 600) 3 500 (2 100, 5 000) 5 300 (160, 10 000) 7 900 000 (6 700 000, 9 100 000) 350 000 (140 000, 550 000) 1 300 000 (1 200 000, 1 500 000) 690 000 (37 000, 3 700 000) 16 000 (−520, 31 000)

62 (27, 96) 290 (210, 360) 530 (150, 870) 190 (0, 580) 430 000 (220 000, 640 000) 140 000 (63 000, 200 000)

wildfires

14 000 (10 000, 18 000) 36 000 (20 000, 53 000) 16 000 (5 800, 25 000) 4 800 (2 500, 8 200) 5 200 (3 100, 7 300) 7 800 (240, 15 000) 11 000 000 (9 400 000, 13 000 000) 480 000 (190 000, 760 000) 1 900 000 (1 600 000, 2 100 000) 940 000 (51 000, 5 000 000) 21 000 (−710, 43 000)

3 300 (1 500, 5 200) 15 000 (11 000, 29 000) 29 000 (8 200, 48 000) 11,000 (2, 32 000) 21 000 000 (11 000 000, 31 000 000) 7 000 000 (3 100 000, 9 800 000)

international

17 000 (12 000, 21 000) 43 000 (23 000, 62 000) 19 000 (6 900, 30 000) 5 700 (3 000, 9 800) 6 400 (3 800, 9 000) 9 000 (280, 18 000) 13 000 000 (11 000 000, 15 000 000) 560 000 (230 000, 890 000) 2 200 000 (1 900 000, 2 400 000) 1 100 000 (60 000, 5 800 000) 25 000 (−850, 50 000)

450 (200, 710) 2 100 (1 500, 2 600) 4 200 (1 100, 6 900) 1,500 (0, 4 400) 2 800 000 (1 400 000, 4 200 000) 930 000 (420 000, 1 300 000)

secondary organics & biogenics

National Morbidity Mortality Air Pollution Study (Bell et al. 2004). bMeta-analysis of short-term ozone mortality studies (Levy et al. 2005). cAmerican Cancer Society cohort study (Krewski et al., 2009). Harvard Six Cities cohort study (Laden et al., 2006).

d

a

acute bronchitis

asthma exacerbation

work loss days

upper & lower respiratory symptoms

acute respiratory symptomsd

respiratory ED visitss

respiratory hospital admissions

cardiovascular hospital admissions

acute myocardial infarctions

H6Cd

PM2.5-related impacts premature death ACSc

lost school days

acute respiratory symptoms

respiratory ED visits

respiratory hospital visits

meta-analysisb

ozone-related impacts premature death multi-citya

emission sector

Table 1. Ozone and PM2.5-Related Premature Deaths and Illnesses Attributable to Each of 7 Emission Sectors in 2016

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

Figure 4. Percentage of annual all-cause deaths due to PM2.5 and ozone precursors from the electricity generating and mobile source sectors in 2005 and 2016 in U.S. counties.

bias = −9%), although it is clear from SI Table 4 and Figure 5 that some of this good performance for total PM2.5 is the result of over and under-predictions of the speciated components. This is further evidence that only providing PM2.5 performance metrics is not a good indicator of model performance; validation must be done for the component species. The modeling system does well at capturing the month to month variability in observed components of PM2.5 (SI Figure 6). Over all 1102 ozone monitors used in this analysis (SI Figure 7), average ozone season (May through September for this analysis) 8 h maximum ozone is well characterized by the modeling system. The mean observed value is 51.0 ppb and the mean modeled value is 51.8 ppb. Over all sites, there is a mean bias of 0.8 ppb and mean error of 9 ppb (fractional bias = 3% and fractional error = 18%). Model estimates of ozone compare well with observation data collected during 2005 (SI Figure 8), giving us confidence that the modeling system is appropriately characterizing the complex nature of precursor emissions and

secondary formation processes. Similarly to speciated PM2.5, the modeling system does well at capturing the seasonal trends in ozone (SI Figure 8). Contribution to PM2.5 and O3 Air Quality. We estimated contributions from each sector tracked in this assessment at all PM2.5 federal reference method and ozone monitors (SI). Figure 2 summarizes: the annual average PM2.5 contribution from the sectors analyzed in this study as a distribution over all FRM monitors for both the 2005 and 2016 simulations and to daily 8 h maximum ozone concentration at all ozone monitors.. For annual average PM2.5, the biggest contributors (on average) are EGU point sources, nonpoint area, mobile (onroad and nonroad), and residential wood combustion. The largest contributing anthropogenic sectors to ozone concentrations in 2005 are mobile sectors: onroad, nonroad, and air/ rail/marine. Between 2005 and 2016, large decreases in contribution from the onroad sector to ozone and to a lesser degree to annual PM2.5 are evident, which reflects a significant 3585

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

2005 and 2016 the population-weighted exposure to directly emitted PM2.5 and particulate sulfate each decline, while exposure to particulate nitrate increases modestly. The significant decline in overall particulate sulfate exposure is driven largely by reductions in the EGU and mobile sectors. The fall in direct PM2.5 exposure, and the small rise in particulate nitrate exposure, are each attributable to the mobile sector. Temporal and Spatial Distribution of Premature Deaths from Anthropogenic Emissions. When characterizing the change in health impacts over time and space, we focus on changes in the incidence of premature death. Comparing mortality results for 2005 with 2016, we find that the number of premature deaths attributable to EGUs and mobile sources decreases significantly over time in the modeled scenario (by about 53% and 36%, respectively; Figure 3). The number of premature deaths associated with emissions from industrial sources show a modest decline (about 6%). Conversely, area sources, and international sources, show small increases. Increases in premature deaths from these latter sectors are attributable to increases in the number of individuals exposed rather than increases in emissions from these sectors. For the mobile source sector, the overall magnitude of the PM and ozone mortality risk, and the geographic scope of that risk, decline between 2005 and 2016 (Figure 4). In 2005, we estimate that the percentage of all deaths attributable to PM2.5 and ozone from mobile source emissions is at or above 1% in large portions of the Midwest, Northeast, and California, and in some areas of Texas and the Pacific Northwest. By 2016, the number of counties at or above this level declines significantly; very few counties in either Texas or the Pacific Northwest exceed this level. The percentage of total deaths attributable to PM2.5 and ozone due to EGUs also declines significantly. In 2005, the percentage of deaths due to PM2.5 and ozone precursors originating from this sector is at or above 1.7% among most counties in the Eastern U.S. In the 2016 scenario, we estimate that no Eastern counties are at or above this level, and most counties are at or below 1.1%.

decline in emissions from this sector in the modeled scenario due to the greater market penetration of lower emitting vehicles in the future and implementation of Federal control programs. There is a similarly large drop in the contribution from the EGU sector to annual (SI) PM2.5, which is consistent with emissions reductions expected as part of planned Federal emissions reductions for this sector. Lateral boundary inflow is a notable source group in terms of contribution for ozone and PM2.5. While not as well characterized as U.S. anthropogenic sources, the portions of Canada and Mexico in the model domain are tracked for contribution and have notable contributions to areas inside the U.S. with close proximity to the border. Health Impacts Attributable to Each Sector from Anthropogenic Emissions in 2005 and 2016. For 2005, we estimate that EGU emissions contribute most to annual PM2.5-related health impacts (38 000 premature deaths, 90% confidence intervals from 27 000 to 48 000, and 620 000 life years lost), followed by area sources (19 000 premature deaths, 90% confidence intervals from 14 000 to 24 000, and 310 000 life years lost) (Figure 3 and SI). We estimate that 2005 emissions from the mobile sector emissions contribute the second largest number of PM2.5 and ozone-related health impacts (Figure 3 and SI). For each emission sector modeled, the number of PM2.5-related deaths is greater than the number of ozone-related premature deaths, with the exception of biogenic emissions, where the ozone-related deaths are slightly greater in both 2005 and 2016 (SI). As noted above, these estimates are sensitive to our assumptions in our model emissions projections regarding the timing and magnitude of emission reductions from each sector. As discussed in detail below, we lack projected emissions inventories for certain anthropogenic emission sectors, including residential wood combustion, and so the projected health impacts for these sources account for changes in population growth but not emissions. We also estimated the number of cases of PM2.5-related acute myocardial infarctions, cardiovascular and respiratory hospital admissions, respiratory emergency department visits, acute respiratory symptoms, upper and lower respiratory symptoms, lost work days, cases of asthma exacerbation and acute bronchitis. For ozone, we quantified the number of respiratory hospital admissions, lost school days, and acute respiratory symptoms. For ease of presentation, we aggregate these morbidity impacts among seven groups of sectors in 2016: (1) industrial point sources; (2) area sources; (3) EGUs; (4) mobile sources; (5) wildfires; (6) international emissions; and, (7) secondary organic, and biogenic emissions (Table 1). For 2016, we estimate the greatest number of PM2.5 and ozone morbidity impacts among the area sector group, where we quantify 33 000 acute myocardial infarctions, 45 000 total cardiovascular and respiratory hospital admissions and emergency departments, 1.1 million upper and lower respiratory symptoms, 480 000 cases of acute bronchitis, 2.1 million cases of asthma exacerbation, 4.3 million lost work days, and 28 million cases of acute respiratory symptoms. To provide insight to the PM species that are associated with each of these broad source groupings we report the populationweighted exposure to directly emitted PM2.5, particulate nitrate and particulate sulfate in 2005 and 2016 for each of seven broad sector definitions (SI). To the extent that certain of these species are more toxic than others, then our estimates of health impacts for each sector may be under- or overstated. Between

■

DISCUSSION Using photochemical source apportionment modeling and health impact functions, we estimated the public health burden of PM2.5 and ozone that is attributable to each of emission sectors in 2005 and 2016. Applying 2005 air quality estimates, we quantify tens of thousands of premature deaths, hospital admissions, non-fatal heart attacks and other acute impacts due to PM2.5 and/or ozone precursor emissions from industrial point, EGU, mobile, international and secondary organic aerosols, and thousands of deaths from other sources including industrial point sources, wildfires, and residential wood combustion among others (SI). We estimate thousands of PM2.5 and ozone-related life years lost due to emissions from these same sources. The level of PM2.5 and ozone mortality risk drops markedly for EGUs and mobile sources between 2005 and 2016, such that the mortality impacts these sectors pose declines from the first and second largest among all sectors to third and fourth (Figure 3). The reduction in mortality risk from EGUs and mobile sources reflects the implementation of various rules affecting the emissions of directly emitted PM2.5 and PM2.5 and ozone precursors from these sectors between 2005 and 2016. The Mercury and Air Toxics Standards, not accounted for in this analysis, promise to further reduce the mortality burden 3586

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

Third, this modeling does not perturb the atmosphere by changing the relative levels of PM2.5 and ozone precursors, and so the health burden we estimate is associated with the fraction of total PM2.5 and ozone levels attributable to each sector at a given point in time (2005 or 2016), holding all else constant. Thus, the health benefits resulting from emission controls placed on a given sector are likely to differ from those reported here and will depend on the relative levels of other precursors and the level of emission reductions. Further, uncertainties associated with the emissions or meteorological inputs may lead to deficiencies in the characterization of the source contribution mix at a receptor location. Our approach used a complete year of meteorology to capture the variety of ozone formation regimes. However, it is possible that the meteorology used for these model applications may not represent all ozone formation regimes at every individual receptor location in the continental United States. Finally, this analysis assumes that all fine particles are equally toxic. We acknowledge above that a growing body of literature suggests that variability in the size of risk estimates may be explained in part by both particle composition, season, and source type. However, we argue that this literature does not yet provide the data necessary to estimate PM-related impacts that are differentiated by specie. The population-weighted estimates of exposure to each PM species above confirm that reductions in total PM2.5 levels have a nonuniform affect on particle composition. And, to the extent that certain species are more toxic than others, then this assessment may have under- or overestimated the PM health impacts attributable to each sector. And, to the extent that certain sectors affect PM levels differently across seasons, and risks also differ across seasons, then the impacts reported here may be under- or overestimated. Our analysis suggests that emissions from EGUs and area sources contribute more than other emission source sectors to air pollution-related health impacts in the U.S. in 2005. We estimate that the overall burden of outdoor air pollution on mortality will decline from 2005 to 2016, particularly for EGUs and mobile sources due to implementation of regulatory programs affecting emissions from these sources.. Notwithstanding the limitations and uncertainties noted above, our results indicate that while that overall levels of PM2.5 and ozone are improving, many sectors continue to pose a considerable burden to public health. Moreover, even if the emissions and air quality impacts attributable to these sectors were to remain constant, population growth would likely increase the total number of premature PM2.5 and ozone-related deaths in the future.

from this sector. The continuing shift to newer, lower emitting cars (i.e., vehicle fleet turnover) and rules including Tier-2 vehicle standards, the on-road and nonroad rules, among others, will reduce emissions of PM2.5 and ozone precursors between 2005 and 2016; indeed, continued turnover in the vehicle fleet promises to further reduce emissions well beyond 2016. The changes in mortality impacts for other sectors are more muted. The reduction in premature mortality from industrial point sources is a result of National Emission Standards for Hazardous Air Pollutants, whose emission control requirements for air toxics frequently reduce emissions of PM2.5 and ozone precursors as a cobenefit. As we note above, the increases in premature deaths attributable to certain sectors including area sources, international emissions, secondary organic aerosols, biogenic emissions, and wildfires are due largely to increases in population growth, rather than growth in emissions from those sectors. The heterogeneity in air quality and health impacts estimated in this paper argues for the importance of tracking high resolution sectors with source contribution techniques to provide more specificity of culpability beyond very broadly defined sectors such as point, area, and mobile. This analysis is subject to important limitations and uncertainties, some of which are common to quantitative health risk assessments, but certain of which are unique to this assessment. For example, uncertainties associated with predicting recent and future total air quality levels, transferring risk coefficients from the epidemiological literature, projecting the growth and distribution of the population, are each described in detail elsewhere.47 However, we believe that some of these uncertainties and limitations have the potential to influence our analysis greatly and are worth noting here. First, our analysis does not reflect the substantial improvements in PM2.5, and to a lesser extent ozone, air quality levels expected to result from the implementation of the Mercury and Air Toxics Standards. The Regulatory Impact Analysis accompanying that rule estimated about 4200 fewer PM2.5related premature deaths as a cobenefit of emission controls reducing releases of mercury from EGUs in 2016. These results suggest that the overall contribution of PM2.5-related mortality attributable to EGUs in 2016 is somewhat lower than estimated in this paper and so the decline in premature deaths would be steeper for this sector between 2005 and 2016. Moreover, this assessment does not reflect reductions in PM2.5 and ozone expected from the full implementation of emission controls for mobile sources due to fleet turnover expected beyond 2016. Finally, our analysis does not account for human health risks from air toxics, and so it underestimates the overall magnitude of health risks attributable to recent and projected levels of air pollution. Second, when projecting the emission reductions expected to result from the implementation of these programs, we used the best available information. However, the timing and magnitude of emission reductions from these programs may differ from our modeling in ways that could affect the size and distribution of health impacts attributable to each sector. Moreover, we have greater confidence in the emissions projections for some sectors than others. For example, the emissions for electrical generating units are projected using IPM, which characterizes well the unit-level emissions of PM2.5 and ozone precursors. Conversely, we estimate emissions from the residential wood combustion sector indirectly, using population surrogates, and resolve these emissions at the county level.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information accompanying this article contains additional results and information regarding our approach for estimating population exposure; the health impact functions we applied; and the model performance evaluation. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*Phone: (919) 541-0209; fax: (919) 541-0839; e-mail: Fann. [email protected]. 3587

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

Notes

(18) Grabow, M. L.; Spak, S. N.; Holloway, T.; Jr, B. S.; Mednick, A. C.; Patz, J. A. Environmental Health Perspect. Population (English Edition) 2011. (19) Regulatory Impact Analysis for Revisions to the Particulate Matter National Ambient Air Quality Standards; U.S. Environmental Protection Agency: Research Triangle Park, NC, 2012. (20) Health Benefits of the Second Section 812 Prospective Study of the Clean Air Act, Review of EPA’s DRAFT; U.S. EPA Science Advisory Board: Washington, DC, 2010. (21) Yarwood, G.; Rao, S.; Yocke, M.; Whitten, G. Updates to the Carbon Bond Mechanism: CB05; Yocke and Company: Novato, CA, 2005. (22) Chang, J. S.; Brost, R. A.; Isaksen, I. S. A.; Madronich, S.; Middleton, P.; Stockwell, W. R.; Walcek, C. J. A three-dimensional Eulerian acid deposition model: Physical concepts and formulation. J. Geophys. Res. 1987, 92, 14681. (23) Nenes, A.; Pandis, S. N.; Pilinis, C. Continued development and testing of a new thermodynamic aerosol module for urban and regional air quality models. Atmos. Environ. 1999, 33, 1553−1560. (24) Strader, R. Evaluation of secondary organic aerosol formation in winter. Atmos. Environ. 1999, 33, 4849−4863. (25) Baker, K.; Scheff, P. Photochemical model performance for PM2.5 sulfate, nitrate, ammonium, and precursor species SO2, HNO3, and NH3 at background monitor locations in the central and eastern United States. Atmos. Environ. 2007, 41, 6185−6195. (26) Nobel, C. E.; McDonald-Buller, E. C.; Kimura, Y.; Allen, D. T. Accounting for spatial variation of ozone productivity in NOx emission trading. Environ. Sci. Technol. 2001, 35, 4397−4407. (27) Russell, A. G. EPA Supersites Program-related emissions-based particulate matter modeling: Initial applications and advances. J. Air Waste Manage. Assoc. 2008, 58, 289−302. (28) Bey, I.; Jacob, D. J.; Yantosca, R. M.; Logan, J. A.; Field, B. D.; Fiore, A. M.; Li, Q.; Liu, H. Y.; Mickley, L. J.; Schultz, M. G. Global modeling of tropospheric chemistry with assimilated meteorology: Model description and evaluation. J. Geophys. Res. 2001, 106, 23073. (29) U.S. EPA. 2005 National Emissions Inventory Data & Documentation, 2011; http://www.epa.gov/ttnchie1/net/ 2005inventory.html. (30) U.S. EPA. North American Emissions InventoriesMexico, 2011. http://www.epa.gov/ttnchie1/net/mexico.html. (31) U.S. EPA North American Emissions InventoriesCanada; U.S. EPA: Research Triangle Park, NC, 2011. (32) U.S. EPA. Integrated Planning Model Analysis for Mercury and Air Toxics Standards. http://www.epa.gov/airmarkets/progsregs/epa-ipm/ toxics.html (accessed November 1, 2012). (33) U.S. EPA. Regulatory Impact Analysis for the Proposed Federal Transport Rule,Docket ID No. , EPA-HQ-OAR-2009-04912010. http://www.epa.gov/ttn/ecas/regdata/RIAs/proposaltrria_final.pdf. (34) U.S. EPA. Regulatory Impact Analysis: National Emissions Standards for Hazardous Air Pollutants for Industrial, Commercial, and Institutional Boilers and Process Heaters, 2011. http://www.epa.gov/ ttn/ecas/regdata/RIAs/boilersriafinal110221_psg.pdf. (35) U.S. EPA. Regulatory Impact Analysis (RIA) for Existing Stationary Spark Ignition (SI) RICE NESHAP, EPA-452/R-10-010, 2010. http://www.epa.gov/ttn/ecas/regdata/RIAs/riceriafinal.pdf. (36) US EPA Regulatory Impact Analysis for the NOx SIP Call, FIP, and Section 126 Petitions, EPA-452/R-98-003, 1998. http://www.epa. gov/ttn/oarpg/otag/sipriav2.zip. (37) Regulatory Impact Assessment for the Mercury and Air Toxics Standards; U.S. Environmental Protection Agency, 2011. (38) Baker, K. R.; Foley, K. M. A nonlinear regression model estimating single source concentrations of primary and secondarily formed PM2.5. Atmos. Environ. 2011, 45, 3758−3767. (39) Wagstrom, K. M.; Pandis, S. N.; Yarwood, G.; Wilson, G. M.; Morris, R. E. Development and application of a computationally efficient particulate matter apportionment algorithm in a threedimensional chemical transport model. Atmos. Environ. 2008, 42, 5650−5659.

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We gratefully acknowledge the assistance we received from both Susan C. Anenberg and Bryan J. Hubbell, whose feedback greatly improved early versions of this manuscript.

■

REFERENCES

(1) Air Quality Criteria for Ozone and Related Photochemical Oxidants; U.S. Environmental Protection Agency: Research Triangle Park, NC, 2006. (2) US EPA Integrated Science Assessment for Particulate Matter (Final Report); U.S. Environmental Protection Agency: Research Triangle Park, NC, 2009. (3) Pope, C. A.; Dockery, D. W. Health effects of fine particulate air pollution: Lines that connect. J. Air Waste Manage. Assoc. (1995) 2006, 56, 709−42. (4) Bell, M. L.; McDermott, A.; Zeger, S. L.; Samet, J. M.; Dominici, F. Ozone and Short-term Mortality in 95 US Urban Communities, 1987−2000. JAMA, J. Am. Med. Assoc. 2004, 292, 2372−2378. (5) Pope, C. A.; Muhlestein, J. B.; May, H. T.; Renlund, D. G.; Anderson, J. L.; Horne, B. D. Ischemic heart disease events triggered by short-term exposure to fine particulate air pollution. Circulation 2006, 114, 2443−2448. (6) Ostro, B.; Lipsett, M.; Mann, J.; Braxton-Owens, H.; White, M. Aair pollution and exacerbation of asthma in African-American children in Los Angeles. Epidemiology 2001, 12, 200−208. (7) Krewski, D.; Jerrett, M.; Burnett, R. T.; Ma, R.; Hughes, E.; Shi, Y.; Turner, M. C.; Pope, C. A.; Thurston, G.; Calle, E. E.; Thun, M. J.; Beckerman, B.; DeLuca, P.; Finkelstein, N.; Ito, K.; Moore, D. K.; Newbold, K. B.; Ramsay, T.; Ross, Z.; Shin, H.; Tempalski, B. Extended follow-up and spatial analysis of the American Cancer Society study linking particulate air pollution and mortality. Research report (Health Effects Institute) 2009, 5−114 discussion 115−36.. (8) Anenberg, S. C.; Horowitz, L. W.; Tong, D. Q.; West, J. J. An estimate of the global burden of anthropogenic ozone and fine particulate matter on premature human mortality using atmospheric modeling. Environ. Health Perspect. 2010, 118, 1189−95. (9) CohenA. J.AndersonH. R.OstroB.Pandey, K. D.; Krzyzanowski, M.; Künzli, N. Urban air pollution. In Comparative Quantification of Health Risks: Global and Regional Burden of Disease Due to Selected Major Risk Factors; World Health Organization, 2004; pp 1353−1434 (10) U.S. Environmental Protection Agency Quantitative Health Risk Assessment for Particulate Matter; Research Triangle Park, NC, 2010. (11) Fann, N.; Lamson, A. D.; Anenberg, S. C.; Wesson, K.; Risley, D.; Hubbell, B. J. Estimating the National Public Health Burden Associated with Exposure to Ambient PM2.5 and Ozone. Risk Anal. 2011, 32, 81−95. (12) ENVIRON User’s Guide Comprehensive Air Quality Model with Extensions version 5.30, 2010. www.camx.com. (13) Environmental Benefits Mapping and Analysis Program (BenMAP); Abt Associates, 2010. (14) Levy, J. I.; Baxter, L. K.; Schwartz, J. Uncertainty and variability in health-related damages from coal-fired power plants in the United States. Risk Anal. 2009, 29, 1000−1014. (15) Hubbell, B. J.; Hallberg, A.; McCubbin, D. R.; Post, E. Healthrelated benefits of attaining the 8-hr ozone standard. Environ. Health Perspect. 2004, 113, 73−82. (16) Voorhees, A. S.; Fann, N.; Fulcher, C.; Dolwick, P.; Hubbell, B.; Bierwagen, B.; Morefield, P. Climate change-related temperature impacts on warm season heat mortality: A proof-of-concept methodology using BenMAP. Environ. Sci. Technol. 2011, 45, 1450−1457. (17) Fann, N.; Fulcher, C.; Hubbell, B. The influence of location, source, and emission type in estimates of the human health benefits of reducing a ton of air pollution. Air Qual., Atmos. Health 2009, 2, 169− 176. 3588

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589

Environmental Science & Technology

Policy Analysis

(40) Wagstrom, K. M.; Pandis, S. N. Source-receptor relationships for fine particulate matter concentrations in the Eastern United States. Atmos. Environ. 2011, 45, 347−356. (41) Kwok, R.; Fung, J. C. H.; Lau, A. K. H.; Wang, Z. S. Tracking emission sources of sulfur and elemental carbon in Hong Kong/Pearl River Delta region. J. Atmos. Chem. 2012, 69, 1−22. (42) University of North Carolina Institute for the Environment Source Apportionment Peer Review. http://www.epa.gov/scram001/ reports/SourceApportionmentPeerReview.pdf (accessed March 3, 2013). (43) U.S. EPA Modeled Attainment Test Software User’s Manual, 2010. http://www.epa.gov/ttn/scram/guidance/guide/MATS-2-3-1_ manual.pdf. (44) Frank, N. H. Retained nitrate, hydrated sulfates, and carbonaceous mass in Federal Reference Method fine particulate matter for six eastern U.S. cities. J. Air Waste Manage. Assoc. 2006, 56, 500−511. (45) USEPA. Guidance on the Use of Models and Other Analyses for Demonstrating Attainment of Air Quality Goals for Ozone, PM2.5, and Regional Haze, RTP, OAQPS; U.S. Environmental Protection Agency, 2007; Vol. EPA-454/B-. (46) Simon, H.; Baker, K. R.; Phillips, S. Compilation and interpretation of photochemical model performance statistics published between 2006 and 2012. Atmos. Environ. 2012, 61, 124−139. (47) Roman, H. a; Walker, K. D.; Walsh, T. L.; Conner, L.; Richmond, H. M.; Hubbell, B. J.; Kinney, P. L. Expert judgment assessment of the mortality impact of changes in ambient fine particulate matter in the U.S. Environ. Sci. Technol. 2008, 42, 2268−74.

3589

dx.doi.org/10.1021/es304831q | Environ. Sci. Technol. 2013, 47, 3580−3589