Public Health and Climate Benefits and Trade‐Offs of U.S. Vehicle Electrification

Abstract Vehicle electrification is a common climate change mitigation strategy, with policymakers invoking co‐beneficial reductions in carbon dioxide (CO2) and air pollutant emissions. However, while previous studies of U.S. electric vehicle (EV) adoption consistently predict CO2 mitigation benefits, air quality outcomes are equivocal and depend on policies assessed and experimental parameters. We analyze climate and health co‐benefits and trade‐offs of six U.S. EV adoption scenarios: 25% or 75% replacement of conventional internal combustion engine vehicles, each under three different EV‐charging energy generation scenarios. We transfer emissions from tailpipe to power generation plant, simulate interactions of atmospheric chemistry and meteorology using the GFDL‐AM4 chemistry climate model, and assess health consequences and uncertainties using the U.S. Environmental Protection Agency Benefits Mapping Analysis Program Community Edition (BenMAP‐CE). We find that 25% U.S. EV adoption, with added energy demand sourced from the present‐day electric grid, annually results in a ~242 M ton reduction in CO2 emissions, 437 deaths avoided due to PM2.5 reductions (95% CI: 295, 578), and 98 deaths avoided due to lesser ozone formation (95% CI: 33, 162). Despite some regions experiencing adverse health outcomes, ~$16.8B in damages avoided are predicted. Peak CO2 reductions and health benefits occur with 75% EV adoption and increased emission‐free energy sources (~$70B in damages avoided). When charging‐electricity from aggressive EV adoption is combustion‐only, adverse health outcomes increase substantially, highlighting the importance of low‐to‐zero emission power generation for greater realization of health co‐benefits. Our results provide a more nuanced understanding of the transportation sector's climate change mitigation‐health impact relationship.


Introduction
Emission and accumulation of greenhouse gases (GHGs) in Earth's atmosphere has increased radiative forcing, led to global climatic change, and motivated mitigation and adaptation planning (Intergovernmental Panel on Climate Change, 2018; Myhre et al., 2014). Among the most compelling GHG reduction measures proposed are those with economic, social, and/or health co-benefits (Haines, 2017;Patz et al., 2020;Rogelj et al., 2016;Thompson et al., 2014). Co-beneficial actions include relatively facile measures, such as the promotion of active transport and reduced meat consumption, as well as more high-inertia efforts including preservation and expansion of urban greenspaces and the transformation of carbon-intense transportation and energy infrastructure. Evidence suggests that quantifying co-beneficial outcomes in climate change mitigation policy analyses directly addresses concerns of political leaders-primarily that CO 2 mitigation is costly and has limited local benefits (Granoff et al., 2016;Nemet et al., 2010). Indeed, previous efforts seeking to elucidate the ancillary benefits of GHG reduction scenarios have demonstrated that the economic benefits from concomitant reductions in air pollutants and their attendant health impacts can exceed the costs of GHG abatement (West et al., 2013). Here, we present a transportation sector-targeted analysis of the co-benefits and trade-offs of the electrification of light-duty passenger vehicles (LDPV) in the United States. EV adoption has the potential to provide concomitant reductions in air pollutants and GHGs thereby providing both positive (and often local) health and economic benefits.
We target the U.S. light-duty transportation sector for several reasons. In 2017, U.S. transportation sector GHG emissions surpassed all other individual sectors, accounting for 29% of the country's total GHG emissions. Within the transportation sector,~60% of GHG emissions came from light-duty vehicles (U.S. Environmental Protection Agency [EPA], 2019a). As governments at the state-, county-, and city-level develop Climate Action Plans (CAPs) to reduce GHG contributions, transportation GHG reductions have been a key focus (City of Chicago, 2008; Los Angeles County Department of Regional Planning, 2015; New York City Mayor's Office of Sustainability, 2017). Additionally, nine states have identified light-duty passenger vehicles as their single largest GHG emissions source and implemented an action plan to accelerate electric vehicle adoption through strategies including infrastructure investment and consumer incentives (Multi-State Zero Emission Vehicle (ZEV) Task Force, 2018).
In addition to being a leading contributor to GHG emissions, the U.S. transportation sector is responsible for air pollutant emissions that cause a substantial public health burden. Light-duty vehicle emissions include primary and secondary pollutants that comprise or contribute to atmospheric fine particulate matter (PM 2.5 ) and ground-level ozone (O 3 )-both of which are criteria air pollutants with well-documented human health impacts (U.S. EPA, 2018a). Fann et al. (2013) estimate that ground-level O 3 and PM 2.5 from mobile source emissions cause between 19,300 and 54,000 premature deaths per year in the United States. Similarly, a recent International Council on Clean Transportation report estimated that the United States experienced 22,000 transportation-attributable ambient PM 2.5 and O 3 deaths in 2015 (Anenberg et al., 2019). Most recently, Davidson et al. (2020) estimated a health burden of 12,000-31,000 premature deaths in the United States for on-road emissions alone in the year 2011. Given the magnitude of the health burden associated with ambient air pollution from traffic, reducing vehicle emissions through vehicle electrification is a clear opportunity for mitigating air pollution-related health effects while also reducing climate forcing from CO 2 and short-lived climate pollutant emissions. Indeed, the aforementioned CAPs and ZEV Action Plan all cite potential air quality co-benefits of reduced vehicle emissions. Therefore, a comprehensive analysis of climate and health co-benefits of transportation electrification scenarios is needed to ensure that EV policy initiatives achieve optimal intended outcomes, particularly when past studies have demonstrated that EV-derived health benefits are not as clear-cut as CO 2 reductions (Requia et al., 2018;Tessum et al., 2014, and references therein).
Prior analyses of the emission changes associated with vehicle electrification have primarily focused on GHG emission reductions from a climate change mitigation perspective (e.g., Requia et al., 2018;Richardson, 2013). Studies have shown that GHG emissions decrease with LDPV electrification even when the power source for battery charging is coal-fired power plants-due to their higher power generation efficiencies in comparison with distributed gasoline-powered internal combustion engines (Huo et al., 2015;Requia et al., 2018). In contrast, the limited existing literature on the air quality impacts of EV adoption has shown a greater dependency on the battery charging energy generation source (Requia et al., 2018). Two China-based EV adoption studies found conflicting results- Huo et al. (2013) found that the total internal combustion engine fleet replacement with EVs would increase PM emissions, while Liang et al. (2019) found that the air quality benefits from EV adoption would avoid over 17,000 deaths annually in addition to reducing GHGs. U.S.-based studies also find nuanced air quality impact differences from EV adoption. Nopmongcol et al. (2017) found that electrification of 17% of light-duty vehicles could lead to modest but widespread reductions in O 3 and particulate matter, whereas Schnell et al. (2019) demonstrated that for PM 2.5 in particular, EV adoption benefits varied by region and season, and depended largely on the power generation mix used for marginal EV charging. Indeed, Tessum et al. (2014) found that in the United States, the health outcome of a 10% EV adoption depended heavily on the type of energy used to charge the EVs. These results reflect a complex trade-off between transportation emissions and power generation emissions and suggest that the regional energy generation mix used to charge EVs heavily influences air quality and health outcomes.
In this study, we focus on premature mortality as a health endpoint, as it captures a range of cardiovascular and respiratory pathways. We use a suite of health impact functions (HIFs) to assess the epidemiological uncertainties from exposure to PM 2.5 and O 3 (Bell et al., 2004;Ito et al., 2005;Jerrett et al., 2009;Krewski et al., 2009;Laden et al., 2006). To elucidate the co-benefits and trade-offs of EV adoption, we use CO 2 emission data and chemistry-climate model simulated surface concentrations of O 3 and PM 2.5 from two different U.S. EV fractional adoption scenarios under three different battery charging power generation configurations. To provide a more comprehensive understanding of the distribution of U.S. health and climate co-benefits under potential future vehicle and energy scenarios, we analyze public health and CO 2 emission benefits and trade-offs for individual states and geographic regions.

Air Quality Simulations and Emissions Scenarios
EV adoption scenarios were developed for coupled atmospheric chemistry and climate model sensitivity simulations by Schnell et al. (2019). We use the model-simulated hourly surface pollutant (O 3 and PM 2.5 ) abundances from Schnell et al.'s six hypothetical EV adoption scenarios (Table 1), as well as the baseline control run. Two different EV adoption proportions were considered under three different energy generation regimes. Conventional LDPV proportions of 25% and 75% were instantaneously replaced with battery powered electric vehicles. To produce the additional electricity to charge EV batteries, varying levels of emission-free power generation sources (wind, solar, hydro, and nuclear) were considered: that is, r0 (no emission-free sources, i.e., all combustion sources), rC (a state's current grid mix), and r2C (doubles each state's fraction of emission-free generation sources). For rC and r2C, the emission-free power generation fraction is based on an individual state's current emission-free generation capacity, that is, any state whose current energy generation is composed of less than 50% emission-free power, will produce a fraction of its needed electricity from combustion sources in the r2C scenario. When referring to these scenarios throughout the paper, we use the notation eX-rY, where X is the percentage of LDPVs converted to EVs and Y indicates the proportion of energy coming from emission-free grid sources (Table 1).
Emission changes (ΔE) resulting from EV adoption were calculated as where E LDVP are the removed emissions (see Table 2) of LDPVs and E EGU are the added emissions from combustion-fired electric generating units (EGUs). LDPV emissions for 2014 are obtained from U.S. Environmental Protection Agency (EPA) National Emissions Inventory 2014 (U.S. EPA, 2014), and power plant emissions are obtained by multiplying remapped electricity demand required by the newly placed EVs by e-GRID reported power plant emission rates (U.S. EPA, 2017). e-GRID is also used for state-level renewable energy fraction, which we assume is uniform across each state . We assume that the adopted EVs have an efficiency of 0.16 kWh km −1 (similar to a 2020 Nissan Leaf or Tesla Model 3 Standard). The complete methods for modifying the emissions for the EV scenarios can be found in Schnell et al. (2019). National emission changes by EV scenario are shown in Table 2.
Emissions developed for the six scenarios were used to drive global simulations of a prototype version of the Geophysical Fluid Dynamics Laboratory Atmospheric model version 4 (GFDL-AM4) model at 0.5°r esolution for the year 2014, following a 1-year spin-up. In our simulations, the GFDL-AM4 atmospheric chemistry-climate model utilizes observed sea surface temperature and sea ice distribution boundary conditions. Simulations use a 30-min time step and are nudged to 2014 NCEP reanalysis winds to facilitate direct comparison to air quality observations (see Figures S1-S3 in Schnell et al., 2019). The model includes detailed tropospheric and stratospheric gas-phase chemistry and simulates the major components of fine particulate matter (PM) including hydrophilic and hydrophobic black carbon and organic matter, ammonium, sulfate, nitrate, sea salt, mineral dust, and secondary organic aerosols (SOA). Sea salt, mineral dust, and biogenic emissions of isoprene and monoterpenes are calculated interactively. Biogenic SOA is emitted as a 5% yield

10.1029/2020GH000275
of isoprene and monoterpene emissions, and anthropogenic SOA is formed by a 5% yield of the oxidation of the lumped n-butane species by OH. Our baseline scenario (BASE) employs Coupled Model Intercomparison Project Phase 6 (CMIP6) emissions (Hoesly et al., 2017) to include surface anthropogenic, biomass burning, and aircraft components. The BASE scenario was evaluated in Section 3.1 "Model evaluation" and Figures S1-S3 of Schnell et al. (2019), which show seasonal biases and correlations between modeled and observed O 3 and PM 2.5 concentrations across the United States. In short, simulated O 3 concentrations were biased high, while PM 2.5 concentrations were biased low, except in the Southeast.
From the hourly model-calculated surface O 3 (ppb) and PM 2.5 (μg m −3 ) abundances over the contiguous United States, we calculate the maximum daily 8-hr average (MDA8) for O 3 and the 24-hr average for PM 2.5 , which serve as inputs for our health impact analyses. To complement the air quality changes and determine co-benefits, for each scenario we compute CO 2 emission changes using the Schnell et al. (2019) remapping algorithm (Equation 1). For example, in the e25-rC scenario, E LDVP is 25% of the total 1.2 Gt U.S. LDPV CO 2 emissions for 2014 and E EGU is the added CO 2 emissions from EGUs.

Health Impact Calculations
We use the U.S. EPA Environmental Benefits Mapping and Analysis Program-Community Edition v1.5 (BenMAP-CE; U.S. EPA, 2019b) to analyze changes in premature mortality resulting from changes in O 3 and PM 2.5 in each electrification scenario. BenMAP-CE calculates changes in adverse health effects using population data, baseline rates of incidence and prevalence of disease, and HIFs from epidemiological studies that quantify associations between health endpoints and changes in pollutant concentrations. The BenMAP-CE baseline mortality data are derived from 2012 to 2014 Center for Disease Control and Prevention's WONDER database, which projects 5-year intervals using annual adjustment factors based on U.S. Census Bureau projected national mortality rates (U.S. EPA, 2018b). Population information is derived from 2010 U.S. Census block-level data projected to 2050 using economic growth factors (Voorhees et al., 2011). Selected HIFs follow a log-linear relationship (Equation 2) to calculate the change in adverse health effects (Δy, deaths avoided year −1 ) for each grid cell, where y 0 represents the baseline incidence rate of the adverse health effect, ΔAQ represents the change in pollutant concentration, Pop is the population exposed, and β is a coefficient derived from the concentration-response function of a given epidemiological study that estimates the response of a health outcome to a change in pollutant concentration (Sacks et al., 2018).
Most of the results we present utilize HIFs from two widely cited epidemiological studies: Krewski et al. (2009) for PM 2.5 and Bell et al. (2004) for O 3 (Table S1). Following recommended best practices, we also assess the sensitivity of our findings to the chosen HIFs by considering two additional functions: Laden et al. (2006) for PM 2.5 and Ito et al. (2005) for O 3 . These additional HIFs quantify relationships between pollutant concentration and health endpoints similar to Krewski et al. and Bell et al. but differ in that they are derived from different cohorts (i.e., different populations, locations, times, and environmental conditions). A large body of evidence in the epidemiological literature links O 3 exposure to short-term mortality (Fann et al., 2012), which both Bell et al. (2004) and Ito et al. (2005) quantify; however two recent studies identified associations between O 3 exposure and long-term mortality Turner et al., 2016). We therefore also include O 3 HIF results from a third study, Jerrett et al. (2009), to consider the sensitivity of our results to long-term O 3 exposure health impacts. We focus the majority of our health impact analyses on results that use the Krewski et al. and Bell et al. methods, as they represent the most conservative damage estimates of the HIFs employed. Table S1 details the properties of each HIF, including study locations, age groups, health endpoints, and timeframe (i.e., long-term vs. short-term). The two PM 2.5 HIFs and the long-term O 3 HIF apply only to adult populations (Krewski et al. & Jerrett et al.,Laden et al.,, and thus the health outcomes we quantify using these functions are limited to these population fractions and are not representative of the total population burden. Pollutant data from our model simulations were input into BenMAP-CE using the appropriate metric for each HIF for each pollutant (i.e., daily mean for PM 2.5 , MDA8 for O 3 ). The change in concentration of each air pollutant is the difference between the BASE simulation and electrification scenario simulations at each 10.1029/2020GH000275 cell in the model grid. The PM 2.5 HIFs and the long-term O 3 HIF quantify annual health endpoints from an annual average pollution metric. Short-term O 3 HIFs quantify aggregated annual health benefits based on daily 8-hr maximum O 3 values during the warm season. We use the full concentration range of PM 2.5 and O 3 and do not apply a minimum concentration threshold when using the HIFs.
To calculate national statistics, we aggregated grid cell results. Regional statistics were calculated using state-aggregated health results, which were then divided into four U.S. regions (Figures 3a and 3g) based on the U.S. Census Bureau's classification system (U.S. Census Bureau, 2018). Regions vary in population size and demographics, baseline incidence rates, electricity generation sources, and corresponding emissions-all of which influence the outcome of the health impact analyses.
The uncertainty associated with the β coefficient of each HIF utilized is included in the β Standard Error column of Table S1, and this uncertainty is incorporated in health impact calculations using Monte Carlo simulations with Latin Hypercube sampling to produce estimates of outcomes at a range of percentiles based on the uncertainty in the health impact function coefficients (Davidson et al., 2007). We report the 2.5% and 97.5% estimates to capture the 95% confidence interval (CI).
To provide an equal-footing comparison of simulated changes in tonnage of CO 2 emissions and premature deaths avoided, we calculate the monetary value of each change using derived damage metrics, that is, the U.S. social cost of carbon (SCC; Ricke et al., 2018) 2018), who use climate model projections, economic damage estimations, and socioeconomic projections to value the expected economic damages from CO 2 emissions. They find that U.S.-level carbon emissions are valued at $48 (66% CI: 1, 118) per ton. To estimate the monetary value of health outcomes we follow the methods employed in the International Council on Clean Transportation report on the global impacts of transportation pollution, that is, the value of statistical life in the United States is found to be $9.6 M using labor market estimates (Anenberg et al., 2019;Viscusi & Masterman, 2017). One caveat to bear in mind when considering these estimates is that our valuation estimates assume that benefits from O 3 and PM 2.5 reductions, while calculated in isolation, can be added together, thereby ignoring potential interactions between coincident reductions of each pollutant.

Pollutant Change Summary
In aggregate, annual average decreases in national ambient concentrations of O 3 and PM 2.5 are simulated under all EV adoption-energy generation scenarios ( Figure 1). Pollutant concentration changes are generally proportional to the EV adoption proportion considered, that is, compare magnitudes of e25 changes to e75 changes. Ozone decreases are found to be largely pervasive across scenarios and states (Figures 1a-1f), whereas changes in PM 2.5 are heterogeneous (Figures 1g-1l). In all simulations, O 3 decreases are largest in the southeastern United States. PM 2.5 increases tend to be localized and are prevalent in regions that rely heavily on combustion power generation facilities. These changes are particularly apparent in combustion-only power generation scenarios (r0). In addition to spatial heterogeneities, Schnell et al. (2019) found heterogeneous seasonal pollutant changes (not shown). For example, during the winter, PM 2.5 decreased throughout the United States in all scenarios due to decreases in nitrate aerosols, whereas increases in PM 2.5 were primarily simulated in the summer and spring due to increases in more thermally stable sulfate aerosols. Seasonal O 3 changes were muted in magnitude, with the exception of the summer when simulated changes were an order of magnitude greater than those simulated in the spring, autumn, and winter.
In the remainder of this manuscript, we focus on annualized health metrics that are calculated from annual data (PM 2.5 and long-term O 3 HIFs) and daily warm season O 3 data (short-term O 3 HIFs). However, spatiotemporal air pollutant variations, extreme events, and changes in exposure potential due to climatic change are also critical for better understanding human exposure and public health outcomes of near-surface pollutant accumulation (Callahan et al., 2019;Gao et al., 2015;Horton et al., 2014). For a comprehensive discussion of the seasonal/regional air pollutant changes and the underlying atmospheric chemistry of the EV adoption scenarios considered here, please consult Schnell et al. (2019).

Aggregated National Climate and Health Co-benefits
Regardless of the EV adoption scenario considered, we find that the United States would achieve aggregate national CO 2 mitigation and mortality reduction benefits (Figure 2a). Aggregate CO 2 mitigation estimates range from 217 Mt year −1 under the least ambitious (e25-r0) adoption scenario wherein EV battery charging is powered by combustion-only sources, to 796 Mt year −1 under the most aggressive adoption-mitigation ; Ricke et al., 2018). Error bars show the 95% CI for health impact results. Circles indicate premature death avoided from changes in O 3 , as calculated using the Bell et al. (2004) HIF. Triangles indicate PM 2.5 premature deaths avoided annually, using Krewski et al. (2009). (b-g) Climate and PM 2.5 health co-benefits and trade-offs  for individual states (smaller circles) and regional averages (larger circles). For population normalized data see Figure S3.
Similar to CO 2 , reductions in ambient O 3 and PM 2.5 concentrations also provide aggregate U.S. benefits for all EV adoption scenarios considered. Air quality improvements due to EV adoption lead to national aggregate decreases in premature mortality for all charging options (Figure 2a and Table 3). Reduced PM 2.5 concentrations decrease annual premature deaths by 358 (95% CI: 242, 473) in e25-r0 to 2,939 (95% CI: 1985, 3,888) in the e75-r2C scenario. Ozone-related premature mortality is likewise reduced in these scenarios, with changes ranging from 96 (95% CI: 32, 159) to 366 (95% CI: 121, 608) in e25-r0 and e75-r2C, respectively. EV adoption under the current energy generation mix leads to PM-related reduced annual mortality that decreases by 437 (95% CI: 295, 578) in e25-rC to 1,576 (95% CI: 1065, 2086) in e75-rC and O 3 -related annual mortality that decreases by 98 (95% CI: 33, 162) to 420 (95% CI: 139, 698), respectively. These reported reduced mortality estimates are obtained from the most conservative dose-response functions we employ (i.e., Bell et al., 2004;Krewski et al., 2009), and comparison with other HIFs (see Table 3 and section 3.3) suggests that the magnitude of avoided mortality could double our estimates. The largest magnitude national-level public health and climate change co-benefits occur in our aggressive adoption scenario when added energy demand for charging is predominantly supplied by emission-free energy sources (e75-r2C). However, in the scenario with the same EV proportion but status-quo energy mix (e75-rC), the magnitude of CO 2 decrease is slight (~9%), while PM 2.5 -related health benefits decreasẽ 46%. The disproportionate influence of the energy mix composition on PM 2.5 has implications for disparate regional health outcomes of vehicle electrification, which we explore further in section 3.2.

Region-and State-Level Co-benefit/Trade-Off Patterns 3.3.1. CO 2 Emission Changes
While we find that aggregate U.S. CO 2 emissions are reduced in all simulated EV scenarios, at the state level there is one exception (Figures 2b and 2e; Table S2). Under the combustion-only (r0) energy generation scenarios, Nebraska experiences CO 2 emission increases of 0.06 Mt year −1 in e25-r0 and 0.18 Mt year −1 in e75-r0 due to a combination of fewer LDPV emissions, high CO 2 emissions from EGUs, and the charging demand of neighboring states. Notably, CO 2 emissions associated with EV charging are tallied where they are produced (i.e., at the power station) and are not necessarily associated with EV charging in that state; that is, a state can see an increase in CO 2 emissions from out-of-state vehicle charging. Indeed, Nebraska experiences CO 2 reductions of 0.68 and 2.0 Mt year −1 respectively for e25-r0 and e75-r0 if only EGU emissions from Nebraska's charging demand is considered. CO 2 mitigation increases moderately with added emission-free energy generation (r0 → rC → r2C), but the total reduction in CO 2 is primarily driven by the replacement of fossil fuel vehicle miles with higher-efficiency energy from power generation stations. For example, California receives the largest CO 2 emission reduction for all EV scenarios, and the statewide change from e25-r0 to e25-r2C reduces CO 2 emissions by~3 M tons, whereas the increased EV proportion from e25-r0 to e75-r0 tripled the CO 2 emission reduction from 29 to 88 M tons per year.

Health Benefits and Trade-Offs From Ozone Changes
Ozone health benefit patterns are similar to CO 2 mitigation benefits, in that they are largely consistent across the six electrification scenarios (Figure 2a). Health benefits from O 3 reduction in individual states in all scenarios are generally positive, relatively modest, and do not exceed 50 fewer premature deaths per year in any  (Bell et al., 2004) and (g-l) PM 2.5 ) drive changes in annual premature mortality incidence. Negative numbers signify increases in premature mortality. Panels (a) and (g) are subdivided into U.S. Census regions: Midwest, West, northeast, and south (U.S. Census Bureau, 2018). For population normalized state and grid cell level data see Figures S1 and S2.

10.1029/2020GH000275
locale (Figures 3a-3f; for population normalized data see Figures S1a-S1f and S2a-S2f)). EV scenarios with combustion-only energy generation still yield O 3 health benefits in most states (Figures 3a and 3d). Under the e25-r0 scenario, wherein EV charging demand is met by combustion-only power generation sources, O 3 decreases, and attendant health benefits are nearly pervasive, however increases in premature mortality are simulated in UT, IA, MN, and PA (Figure 3a). Under the e75-r0 scenario, however, the O 3 decreases and health benefits occur over all states (Figure 3d). Notably, in r2C scenarios with doubled statewide emission-free power generation, isolated states see increased O 3 -driven premature deaths. In e75-r2C the states of New York and New Jersey by and large see O 3 reductions, but O 3 increases in VOC-limited urban grid cells nullify gains in the majority of each state ( Figure S2f). Increases in state-wide premature deaths are minute (NJ: 2.6 year −1 and NY: 0.1 year −1 ), yet positive (Figures S1f and S2f).
By region, the South experiences the largest absolute magnitude of O 3 -related health benefits (Figure 4). For states in the South, median deaths avoided per year range from 56 to 235 across the adoption scenarios (red bars; Figure 4c). In other regions, health benefits due to O 3 reductions are more modest, that is, in all scenarios and all non-South regions cumulative reductions in premature mortality fall below 100 per year. Our

10.1029/2020GH000275
finding that the largest positive health changes are driven by LDPV replacement fraction holds true across regions. Regional median premature deaths avoided per year increase between the e25-rC and e75-rC scenarios from 6 to 42 in the West, 56 to 222 in the South, 14 to 60 in the Northeast, and 22 to 84 in the Midwest. Curiously, we find that in the West, Midwest, and Northeast regions premature mortality reductions decrease as the fraction of emission-free power generation increases, that is, annual deaths avoided in rC scenarios are greater than in r2C scenarios (Figure 4). This counterintuitive result is driven by nonlinear O 3 chemistry in the high population, high emission urban cores of these regions; that is, with sufficient NO x reductions, O 3 production efficiency increases (Sillman, 1995).

Health Benefits and Trade-Offs From PM 2.5 Changes
Unlike CO 2 and O 3 outcomes, the distribution of PM 2.5 -related health consequences is spatially variable and shifts substantially between electrification scenarios (Figures 2b-2g and 3g-3l). In an absolute sense, the most-substantial PM 2.5 health benefits under most scenarios are found in California, Illinois, and the northeastern states of New York, New Jersey, and Pennsylvania ( Figure 3). Population normalized health outcomes suggest a somewhat similar pattern; however on a per capita basis the Midwest and mid-Atlantic states see the largest benefits with emission-free power usage (Figures S1g-S1l, S2g-S2l, and S3a-S3f).
Health outcomes for EV adoption under current energy generation infrastructure vary according to fraction of internal combustion engine replacement. The e25-rC simulation reduces PM 2.5 -related premature mortality in most states, with the exception of the Midwest, Arizona, and New England (Figure 3h). Under the e75-rC scenario, however, all but one state (FL) sees reduced premature mortality (Figure 3k).
When EV adoption uses combustion-only power generation, health outcomes are decidedly more mixed (Figures 3g and 3j). Some northeastern states (e.g., New York and Pennsylvania) consistently experience PM 2.5 -related benefits even under combustion-only charging scenarios due to significant reductions in urban traffic PM 2.5 emissions and precursors and fewer coal-fired power plants in their grid mix. Under e75-r0, we find that 12 states, located mostly in the South and Midwest, experience increases in PM 2.5 -related mortality. This finding is consistent with simulated PM 2.5 increases that result from increases in ammonium sulfate emissions in the summer, driven by SO 2 emissions from coal-fired power plants . Texas, Florida, North Carolina, and Minnesota experience a combined simulated increase of 132 deaths in the e75-r0 scenario (Figure 4j). By contrast, when the same proportion of EVs is charged by higher levels of emission-free energy generation sources (e75-r2C), these four states experience PM 2.5 health benefits totaling 330 avoided premature deaths per year (Figure 4l). Indeed, when aggressive EV adoption is paired with expanded emission-free power generation (e75-r2C) all states see substantial reductions in annual PM 2.5 -related premature mortality. On a regional basis, we find that all sectors have the potential for substantial PM 2.5 -related public health improvements (Figures 2b-2g and 4). The West and Northeast experience net positive benefits of avoided mortality across all scenarios. The Northeast sees the greatest PM 2.5 health benefits of any region in combustion-only (eX-r0) simulations ( Figure 4f) and has larger benefits for e25-r0 than e25-rC, a result of nonlinear NO x chemistry to generate nitrate aerosol ( Figure S10 in Schnell et al., 2019). Under electrification scenarios with doubled fractions of emission-free power generation (eX-r2C), PM 2.5 health benefits in the South exceed other regions, with 394 (95% CI: 266, 521) premature deaths avoided per year in the e25-r2C and 981 (95% CI: 662, 1,298) in e75-r2C (Figure 4g). At 75% EV penetration, the South also has the greatest range of benefits between power generation scenarios; from a predicted increase of 32 premature deaths per year with combustion-only energy sources (e75-r0) to the decrease of 981 deaths per year with doubled emission-free energy sources (e75-r2C). The Midwest also has aggregate increases in adverse health effects under two of the 25% EV penetration scenarios (e25-r0 and e25-rC) with 55 and 18 additional premature deaths per year predicted under these respective scenarios. The PM 2.5 scenario in which each region experiences peak avoided mortality is the most ambitious adoption and electrification scenario (e75-r2C; Figures 2b-2g; Figure S3).

Health Impact Uncertainties
The above reported and discussed health outcomes are based on HIFs from Bell et al. (2004)  However, to elucidate uncertainties in HIFs, for example, differences in mathematical formulation, demographics considered, and temporal exposure (Table S1), we provide national, regional, and state-level comparisons between all HIFs for both O 3 and PM 2.5 exposure (  (Table 3). We also tested the sensitivity of our PM 2.5 health impact results to a third function, Lepeule et al. (2012), which was a follow-up analysis to the Harvard Six Cities study that included data through 2009, that is, an 11-year followup on Laden et al. (2006

Discussion and Conclusion
Our analysis of six EV adoption-energy generation scenarios indicates that vehicle electrification in the United States could annually prevent hundreds-to-thousands of premature deaths while also reducing CO 2 emissions by hundreds of millions of tons. Estimates of economic damages avoided due to EV adoption are substantial. With current infrastructure and 25% EV adoption (e25-rC) we find savings of $16.8B annually (i.e., $11.6B U.S. SCC and $5.1B VSL). In more aggressive scenarios, for example, e75-r2C, savings of $70B year −1 are found (i.e., $38.2B U.S. SCC and $31.7B VSL). However, we also find that PM 2.5 changes and corresponding health impacts vary across U.S. regions, and the realization of co-benefits for PM 2.5 depends largely on the energy sources used to charge EVs. The scenarios that we assess, which assume instantaneous EV replacement of conventional LDV's, are not intended to simulate dynamic real-world EV uptake, but rather serve as sensitivity tests to estimate the magnitude and geographic distribution of mitigation outcomes under varied energy regimes and EV replacement proportions. Additionally, our health impact estimates of avoided mortality and the corresponding economic valuations should not be seen as comprehensive because (a) there is well-documented evidence of a variety of non-mortality health outcomes associated with transportation-related air pollution, including asthma and other cardio-pulmonary ailments, and (b) our study only considers ground-level O 3 and PM 2.5 exposure, while there may be additional impacts from changes in direct exposure to NO 2 and other transport emissions (Anenberg et al., 2018).
Simulated health impact results are influenced by a number of limitations and uncertainties, including HIF cohorts and assumptions, structural chemistry-climate model biases, meteorological variability, and population trends. Our sensitivity analysis using several HIFs demonstrates that relative risk (RR) estimates, and corresponding premature mortality estimates, can vary substantially between epidemiological studies. Our application of HIFs also assumes that the results of cohort studies with specific geographies and populations (Table S1) can be applied nationwide where factors including population demographics, environmental characteristics, and pollutant concentrations will vary from the cohort study conditions. Likewise, the structural formulation of individual chemistry-climate models leads to inherent model response uncertainties, which could ideally be addressed using a multimodel assessment framework (Hawkins & Sutton, 2009). Furthermore, computational demands have limited simulations to a single historic year and therefore do not consider EV adoption and health outcomes in the context of internal climatic variability, a limitation that could be addressed with initial-condition perturbation experiments (Garcia-Menendez et al., 2017). The temporal context of the experiment also presents uncertainties regarding applicability to future years. The meteorology and level of radiative forcing of the modeled year influence pollutant formation, and therefore should be considered when attempting to project future outcomes and applicability (Fiore et al., 2015). Lastly, health impact estimates depend on population demographics and baseline mortality incidence rates, which are expected to shift over time.
National CO 2 mitigation estimates follow a pattern consistent with previous studies that have shown EVs to significantly reduce CO 2 emissions even when the charging energy is sourced from carbon-intensive energy generation facilities including coal combustion (Requia et al., 2018). The range of CO 2 mitigation estimates from our six scenarios represents approximately 18 to 66% of the total 2014 CO 2 emissions attributed to light-duty vehicles in the U.S.-48 (U.S. EPA, 2014). We find that even if the efficiency of the assumed EV was decreased by 30% (i.e., increasing the electricity demand by 30%), CO 2 reductions are still widely apparent-ranging from 192 to 767 Mt year −1 . While these results do not account for the life cycle emissions of EV battery production and disposal, based on past studies we would expect inclusion of battery production emissions to reduce CO 2 mitigation benefits by 5% or less (Samaras & Meisterling, 2008). Further, rapid advances in battery technology, low-carbon manufacturing processes, and the greening of the grid suggest that estimates provided here are likely conservative and that purchased EVs will get cleaner over their lifetimes.
The magnitude of simulated health consequences from PM 2.5 reductions far exceeds the benefits from O 3 changes for most vehicle electrification scenarios. Indeed, the O 3 -attributed national aggregate benefits from our most extreme hypothetical with 75% EV conversion and double emission-free charging (e75-r2C) are comparable with the PM 2.5 -related benefits under the least ambitious of our EV scenarios (e25-r0). The difference in magnitude between PM 2.5 and O 3 -related benefits in our experiments is consistent with literature that indicates the higher relative mortality burden of mobile-source PM 2.5 in the United States compared to O 3 and reflects the higher risk coefficient associated with PM 2.5 exposure compared to O 3 exposure (Fann et al., 2013).
The overall magnitude of PM 2.5 health impact estimates is consistent with most of the limited available literature. Grabow et al. (2012)  Our results demonstrate that PM 2.5 health benefits are not guaranteed under high EV penetration scenarios in some U.S. regions (e.g., Midwest and South), and emission-free energy sources can be the difference between positive and negative state-level health outcomes of vehicle electrification. Given that EV air quality studies (e.g., Ji et al., 2015) have predicted a shift in air pollution burden from urban to rural areas with vehicle electrification-the so-called spillover effect (Fang et al., 2019)-future analyses, ideally employing a higher spatial resolution modeling framework, should investigate the public health consequences of the geographic redistribution of air pollution, which may be an important environmental justice consideration for electric vehicle policy decisions (Ji et al., 2015). This is especially true for urban cores, where the chemical regime (i.e., NO xvs. VOC-limited) and population density may differ substantially over a few tens of kilometers, all of which would be averaged over a single~50 km grid cell in this analysis. Similarly, analyses seeking to project policy outcomes, such as EV adoption, in a future world should strive to characterize all forms of projection uncertainty, that is, internal variability, scenario, and model structure uncertainties, and their implications on projected outcomes (Deser et al., 2020;Kinney, 2008). Studies have projected significant future emissions reductions due to fleet turnover and existing policies such as vehicle fuel and emission standards (Davidson et al., 2020); future work comparing co-benefits from such regulations to vehicle electrification could give insight into the most effective levers to maximize progress toward cleaner air and reduced GHG emissions.
Reducing the environmental burden from the transportation sector is a compelling opportunity to address the two global challenges of air pollution and climate change. In this study, we show the potential distribution of health consequences from vehicle electrification scenarios in the United States and compare the climate CO 2 mitigation outcomes with the health benefits and trade-offs. We find that while U.S. vehicle electrification is expected to significantly reduce transportation CO 2 emissions and has the potential to improve air quality and mitigate thousands of annual premature deaths, the extent and magnitude of health co-benefits largely depend on the charging energy mix, particularly for changes in PM 2.5 . The results show that while electric vehicles under status quo energy regimes produce significant CO 2 reductions, the greatest health co-benefits are achieved by electrifying vehicles and charging with a greater fraction of emission-free electricity generation sources. This finding is consistent with Tessum et al. (2014), who demonstrates the importance of coincident fleet electrification and grid decarbonization to achieve maximum co-benefits. We add to this understanding by elucidating co-benefits and trade-offs on a state-by-state basis and by altering the EV replacement proportion to demonstrate the impact on the magnitude and distribution of climate and health outcomes. Our results reinforce the importance of assessing air quality and health consequences in relation to EV adoption goals. When decision-makers pursue vehicle electrification as a strategy for carbon emission reduction, our results indicate opportunity for widespread and significant public health benefits, but they also show the potential regional trade-offs when EVs are charged by combustion energy generation sources. Policymakers should favor a comprehensive climate action plan that ensures the public will experience optimal health co-benefits in addition to the CO 2 reductions associated with vehicle electrification.