What are the best combinations of fuel-vehicle technologies to mitigate climate change and air pollution effects across the United States?

The transportation sector is the largest contributor to CO2 emissions and a major source of criteria air pollutants in the United States. The impact of climate change and that of air pollution differ in space and time, but spatially-explicit, systematic evaluations of the effectiveness of alternative fuels and advanced vehicle technologies in mitigating both climate change and air pollution are lacking. In this work, we estimate the life cycle monetized damages due to greenhouse gas emissions and criteria air pollutant emissions for different types of passenger-moving vehicles in the United States. We find substantial spatial variability in the monetized damages for all fuel-vehicle technologies studied. None of the fuel-vehicle technologies leads simultaneously to the lowest climate change damages and the lowest air pollution damages across all U.S. counties. Instead, the fuel-vehicle technology that best mitigates climate change in one region is different from that for the best air quality (i.e. the trade-off between decarbonization and air pollution mitigation). For example, for the state of Pennsylvania, battery-electric cars lead to the lowest population-weighted-average climate change damages (a climate change damage of 0.87 cent/mile and an air pollution damage of 1.71 cent/mile). In contrast, gasoline hybrid-electric cars lead to the lowest population-weighted-average air pollution damages (a climate change damage of 0.92 cent/mile and an air pollution damage of 0.77 cent/mile). Vehicle electrification has great potential to reduce climate change damages but may increase air pollution damages substantially in regions with high shares of coal-fired power plants compared to conventional vehicles. However, clean electricity grid could help battery electric vehicles to achieve low damages in both climate change and air pollution.


Introduction
The transportation sector is currently the most significant contributor to CO 2 emissions in the United States (U.S.) (U.S. Energy Information Administration (EIA) 2017). Similarly, the health and environmental consequences associated with the transportation sector are of critical importance, since the transportation sector accounts for more than half of carbon monoxide (CO) and nitrogen oxides (NO x ) emissions in the U.S., as well as nearly a quarter of volatile organic compounds (VOCs), and 6% of primary PM 2.5 (particulate matter less than 2.5 micrometers in diameter) emissions (Davis et al 2016). Increased emissions and concentrations of greenhouse gases and criteria air pollutants are of concern to society and policymakers, as they lead to poor urban air quality, increasing hazards of infrastructure, and elevated risks of mortality and morbidity in exposed populations (Pope et al 2002, Krewski et al 2009, Lepeule et al 2012. However, the social impacts of these pollutants are not the same across pollutant type, space, or time. Greenhouse gases, for example, have global dispersion, stay in the atmosphere for decades to centuries, and their impacts are the same regardless of the location of the source (Intergovernmental Panel on Climate Change (IPCC) 2014). Criteria air pollutants, on the other hand, have a much shorter lifespan, and their consequences depend on the location of the source (Heo et al 2016a, Gilmore et al 2019. Emissions of criteria air pollutants result in higher concentrations of PM 2.5 and ground-level ozone, which lead to substantial health consequences for the exposed population (Heo et al 2016a, Gilmore et al 2019.
The impact of air pollution and that of climate change differ in space (local versus global) and time (short-term versus long-term). This divergent nature of air pollution and climate change has led to different perceptions, rates of technology adoption, and effectiveness of technology options and policy actions to tackle air pollution and climate change (Maione et al 2016, Sergi et al 2018. Whereas efforts to clean the polluted air in the United States and many other developed countries have been hailed as victories (U.S. Environmental Protection Agency (EPA) 2011, Carnell et al 2019), efforts to mitigate climate change have yet to make substantial progress (Peters et al 2017, van Renssen 2018. One promising way to increase the effectiveness of climate change actions is to identify co-benefits (such as health benefit from the reduction of air pollution) which are achieved by technology options or policy actions that are designed initially to tackle climate change (Bain et al 2016, Chang et al 2017, Deng et al 2017. In particular, health co-benefits have been identified for carbon mitigation strategies and policies in the transportation sector (Shindell et al 2011, Balbus et al 2014, Shaw et al 2014, Thompson et al 2014, in the electricity sector (Balbus et al 2014, Thompson et al 2014, Driscoll et al 2015, and across the economy (Zapata et al 2013, Thompson et al 2014, Buonocore et al 2018.
Alternative fuels and advanced vehicle technologies have been widely regarded as a potential solution for sustainable transportation with negligible climate change and air pollution impacts (U.S. National Research Council (NRC) 2010, Creutzig et al 2015, Gearhart 2016. However, quantifications of climate change or air pollution effects of alternative fuels and advanced vehicle technologies are only available recently and limited in scope. In 2010, a U.S. National Academies report estimated that air emissions from on-road vehicles resulted in a cost of~$110 billion for climate change and air pollution damages in the U.S. (U.S. National Research Council (NRC) 2010). Around the same time, several studies have examined climate change damages or air pollution damages caused by alternative fuels and advanced vehicle technologies such as biofuels, compressed natural gas (CNG) vehicles, diesel vehicles, plug-in hybrid electric vehicles, and battery electricity vehicles, and hydrogen electric vehicles (Ogden et Tessum et al 2014, Luk et al 2015, Barrett et al 2015, Tamayao et al 2015, Tong et al 2015a, 2015b, 2017, Weis et al 2015, Yuksel et al 2016, Holland et al 2016a. For a detailed summary of existing literature in terms of fuel-vehicle technologies considered, data sources for emissions inventories, details about air quality modeling (e.g. species, scope, endpoints), and climate change modeling (species and endpoints), we refer the reader to table S1 in the supplementary document (available online at stacks.iop.org/ERL/15/074046/mmedia). In general, these studies find that only a few fuel-vehicle technologies, including hybrid-electric vehicles and battery electric vehicles powered by dedicated natural gas or renewable electricity sources, could reduce climate change damages or air pollution damages compared to incumbent petroleum fuels. However, none of these studies has explicitly quantified the effectiveness of alternative fuels and advanced vehicle technologies in mitigating climate change and air pollution simultaneously in a spatially-explicit manner.
Except for a few recent studies (Hill et al 2009, Tessum et al 2014, Barrett et al 2015, Holland et al 2016a, most existing studies ignored the spatial distributions of air pollution damage. This omission could lead to biased findings because the air pollution impact of criteria air pollutants depends critically on where they are emitted and, subsequently, how many people are exposed. As a result, although many studies estimated and compared air pollution damages and climate change damages for certain fuel-vehicle technologies at the national level (i.e. using a representative estimate), quantification and comparison of air pollution damages and climate change damages at the appropriate spatial context (local versus global) is lacking.
As national and local decision-makers may care about different types of negative consequences, it becomes increasingly important to understand if one particular type or a combination of alternative fuels and advanced vehicle technologies could deliver health benefits as well as climate change benefits compared to the incumbent petroleum fuels. This is particularly relevant since vehicles rely on the refueling or charging infrastructure to support their use (Tong et al 2019). If different alternative fuels or advanced vehicle technologies are favored under different policy goals, it is hard to form joint forces from all the stakeholders and end-use consumers.
To fill this knowledge gap, we present a systematic, spatially-explicit assessment of environmental externalities caused by alternative fuels and advanced vehicle technologies in the United States, using a coupled modeling framework linking life cycle assessment and reduced-form air quality models. We study three typical types of passenger-moving vehicles, passenger cars, SUVs, and transit buses. We include SUVs because their sale has grown to be comparable to that of passenger cars over the last two decades (Davis et al 2016). Although transit buses represent a small share of vehicle fleets, they are essential mobility service providers in urban areas and are early adopters for alternative fuel and advanced vehicle technologies (Tong et al 2017). We include four transportation fuels-gasoline, diesel, CNG, and grid electricitywhich are paired with three vehicle technologiesinternal combustion engine vehicles (ICEVs), hybrid electric vehicles (HEVs), and battery electric vehicles (BEVs)-based on fuel properties and market availability. Table 1 summarizes the life cycle stages for the fuelvehicle technologies considered. The life cycle boundary includes primary energy extraction, fuel production and transportation, and vehicle use. We also include the manufacturing of lithium-ion batteries for HEVs and BEVs following recent literature (Michalek et al 2011, Tessum et al 2014. We assume all other vehicle components are similar across vehicle technologies for a given vehicle type. The functional unit is one vehicle mile traveled (VMT), and the study reference year is 2014 for which we have the complete data.

Marginal damage approach
We use a marginal damage approach to estimate climate change monetized damages associated with greenhouse gases (CO 2 , CH 4 , N 2 O) and health and environmental monetized damages caused by criteria air pollutants (SO 2 , NO x , CO, PM 2.5 , and VOCs). We assume that an incremental VMT is small enough to be treated as marginal so that the induced climate change and air pollution damages are calculated as the product of emissions factors and the marginal damages for the emissions species for the same location and height (U.S. National Research Council (NRC) 2010). The marginal damages of a mass of greenhouse gas are the same regardless of source location and height. However, the marginal damages of criteria air pollutants are sensitive to where they are emitted. We then attribute the monetized costs associated with each fuel-vehicle technology to one vehicle mile. The metric for comparison is monetized damages per vehicle mile. All monetary values are converted to the 2010 dollar using the consumer price index (CPI) (U.S. Bureau of Labor Statistics 2016). We use two state-of-the-art reduced-form air quality models (the AP2 model and the EASIUR model) (Heo et al 2016a, Gilmore et al 2019, Muller 2011 to estimate marginal damages for criteria air pollutants. We present the results with the EASIUR model in the main text and leave the results with the AP2 model in the supplementary document (section 4.4). Critical economic assumptions in our analysis include a statistical life value of $ 2010 8 million and a social cost of carbon of $ 2007 36/t CO 2 e. Given the uncertainty and discussions in the literature regarding these values, we perform a sensitivity analysis to understand the implication of these assumptions on our results.

Greenhouse gas (GHG) emissions and damages
We expand the GHG emissions inventories as reported in Tong et al (Tong et al 2015a(Tong et al , 2015b by considering upstream and on-site GHG emissions for electricity generation for different NERC (North American Electric Reliability Corporation) regions (see the supplementary document for region boundaries and details on the upstream emissions). In particular, a 1.3% methane leakage in the natural gas supply chain was considered (Tong et al 2015a(Tong et al , 2015b. We use the 100-year global warming potential (GWP) from the IPCC (Intergovernmental Panel on Climate Change (IPCC) 2014) to convert non-CO 2 emissions to CO 2 -equivalent emissions (36 for fossil fuel methane and 298 for fossil fuel N 2 O). We use the social cost of carbon ($ 2007 36/t CO 2 e) to monetize the damages associated with GHG emissions (Interagency Working Group on Social Cost of Carbon United States Government 2015). We assess the sensitivity of our results for a range of SCCs from $ 2007 0/t CO 2 to $ 2007 105/t CO 2 .

Criteria air pollutant (CAP) emissions and damages
The life cycle CAP emissions and the resulting damages estimates include those from primary energy extraction, fuel production, vehicle operation, and battery manufacturing.

Primary energy extraction emissions and damages
We calculate the national average air pollution damages producing primary energy products (coal, natural gas, or crude oil) in the U.S. following the approach outlined in (Jaramillo and Muller 2016). We aggregate the county-level criteria air pollutant emissions for these energy sub-sectors from the raw emissions data in the National Emissions Inventory (NEI) (U.S. Environmental Protection Agency (EPA) 2016b) based on the source classification code database developed in (Tschofen et al 2019). We multiply criteria air pollutant emissions and the marginal damage estimates for the same species at the same county to calculate the total damages for energy activities, which are then divided by total energy production to quantify the damages per energy unit. For primary energy extraction, we use national weightedaverage damage in the analysis. Although domestic production of coal and natural gas accounts for the largest share of supply of these fuels, the U.S. continues to import crude oil (U.S. Energy Information Administration (EIA) 2016c) and, like previous studies (Michalek et al 2011, Weis et al 2016, we assume that imported crude oil has the same air pollution damages as U.S. crude oil. We account for air pollution damages due to electricity used for powering crude oil pipelines and natural gas pipelines using pipelines' electricity intensity data (Hooker 1981, Davis et al 2016 and the electricity damages calculated in this paper.

Fuel production emissions and damages
We construct emissions inventories of the four transportation fuels, gasoline, diesel, grid electricity, and CNG, as follows.
We use emissions data from the National Emissions Inventory (NEI) at the county level to estimate air pollution damages associated with the production of gasoline and diesel. These emissions are normalized to oil refinery capacity (barrel of crude oil inputs) before being converted to the actual outputs of petroleum products (gasoline and diesel) using the energy allocation method ( For CNG, we assume compression of natural gas takes place at refueling stations. Thus, compressors use the electricity from the NERC region in which they are located. The energy efficiency of the compression process is 96% (Tong et al 2015a(Tong et al , 2015b. We model GHG and CAP emissions for the U.S. electricity grid in the year 2014. To explore the impact of the evolving electricity grid, we also model a hypothetical case 'alternative electricity grid' in which coal-fired and oil-fired power plants in the 2014 electricity grid are replaced by new natural gas combined cycle power plants at the same location and for the same net generation. For both electricity grids, we use NERC regions as independent geographical units of analysis (U.S. Environmental Protection Agency (EPA) 2015). In the supplementary document, section 2, we include a map of NERC regions for the readers' reference. We account for line losses using U.S. EPA's eGRID data (U.S. Environmental Protection Agency (EPA) 2015), and for battery-electric vehicles, we assume a vehicle charging efficiency of 86.5% (Tong et al 2015a).
To calculate climate change damages and air pollutant damages for electricity generation, we use data from U.S. EPA's Continuous Emissions Monitoring System (CEMS) (U.S. Environmental Protection Agency (EPA) 2016a) (net electricity generation, CO 2 , SO 2 and NO x emissions for fossil fuel power plants) and U.S. EPA's NEI for PM 2.5 emissions. We calculate the air pollution damages from each electric generation units (EGU) by multiplying emissions and their respective county-level marginal damages. To calculate the weighted-average damages associated with the average electricity generation, we sum up the individually-calculated damages for EGUs in each NERC region, which are then divided by the net electricity generation (for the year 2014) in each NERC region. We compile U.S. EIA's Form-923 (U.S. Energy Information Administration (EIA) 2016a) for net generation data for non-fossil fuel power plants, which is not included in U.S. EPA's CEMS.
In the supplementary document, we consider three additional electricity grid scenarios for 2014 electricity grid to understand the likely impacts of electric vehicle charging: (1) the average electricity generation from all fossil fuel power plants ('average fossil-fuel electricity'); (2) marginal fossil fuel electricity that has the lowest combined air pollution and climate change damages for an hour ('cleanest marginal fossil fuel electricity'), and (3) marginal fossil fuel electricity that has the highest combined damages for an hour ('dirtiest marginal fossil fuel electricity').
Data and estimates for net generation, emissions factor, and climate change and air pollution damages associated with electricity generation are available in the supplementary document (section 3.3).

Vehicle operation
Emissions come directly from the combustion of fossil fuels as well as tire and brake wear during vehicle operation. For passenger cars and SUVs, we The vehicle fuel efficiency determines the amount of fuel used per mile. We use the fuel economy assumptions from Tong et al (2015aTong et al ( , 2015b. We also rely on Tong et al (2015aTong et al ( , 2015b for vehicle mileage, battery size, and replacement (for HEVs and BEVs only) assumptions. While outside the scope of this work, other factors, such as temperature and driving profiles, may also affect the overall damages associated with the operation of BEVs as discussed in (Weis et al 2016) and (Yuksel et al 2016).

Battery manufacturing
We use the literature estimates for the air pollution damages associated with battery manufacturing. In particular, the estimated damages between the two literature (Michalek et al 2011, Tessum et al 2014 differed by a factor of three. While the two sources used similar emissions inventories (GREET model), they assumed different locations of production processes and, as a result, found different air pollution damages. We assume an $8.68 kWh −1 damage (the lower estimate) from the more recent source (Tessum et al 2014) in the baseline analysis, and we perform sensitivity analysis with the alternative source (Michalek et al 2011). Battery manufacturing damages are allocated to vehicle miles traveled with the consideration of vehicle lifetime mileage, battery size, and number of batteries per vehicle lifetime (Section 3.4.2 in the supplementary document).

Marginal damages of air pollutants
We use the AP2 model (Muller andMendelsohn 2007, Muller 2011) and the EASIUR model (Heo et al 2016a(Heo et al , 2016b to estimate the marginal damages of air pollutants. Both models use a damage function approach and use similar concentration-response relationships for PM 2.5 (Pope et al 2002, Krewski et al 2009. A key modeling difference is that the AP2 model used a source-receptor matrix framework derived from a Gaussian Plume model, whereas the EASIUR model used a regression method to obtain reduced-form outputs from a tagged chemical transport model (Gilmore et al 2019). The models also differ in baseline emissions inventory, baseline population data, spatial and temporal resolutions, pollutants considered, and damage endpoints assumed (table 2). As a result, there are significant differences in the spatial distributions of their respective marginal damage estimates (see (Heo et al 2016a) and (Gilmore et al 2019) for details). Ongoing work aims to resolve the mentioned differences between the two models.
The EASIUR and AP2 models estimate different marginal damages for ground-level and elevated emissions sources. We use the elevated-level marginal damages for fossil fuel power plants and the ground-level marginal damages for all other emissions sources.
Carbon monoxide is known to affect cardiovascular health and leads to secondary effects from groundlevel ozone (Michalek et al 2011), but neither social cost model includes these effects. Thus, we use the national-average CO damage estimate from (Matthews and Lave 2000) of $520/metric ton CO in $ 1992 , which after inflation adjustment is $ 2010 808/metric ton CO. We

Spatial variability in life cycle damages
We estimate the life cycle climate change damages and air pollution damages for fuel-vehicle technologies used in U.S. counties. In the discussion below, 'damages' or 'life cycle damages' refer to 'life cycle air pollution damages and life cycle climate change damages' , unless otherwise specified. In figure 1, we present the frequency plots (weighted by county population) of county-level damages for each fuel-vehicle technology option considered for passenger cars, SUVs, and transit buses. The estimated life cycle damages show significant heterogeneity across U.S. counties. For instance, the most serious damage is twice as significant as the lowest damage of driving the same gasoline car across counties. However, no matter where the vehicle is driven, the damages are no smaller than 1.5 cents per vehicle mile traveled. For populationweighted-average damages for the Contiguous U.S., gasoline hybrid-electric vehicles lead to the lowest damages for passenger cars and SUVs, whereas battery-electric buses achieve the lowest damages for transit buses (table 3). Weighted-average damage estimates for U.S. states and electricity grid regions are available in the Supplementary Data File.
Hybrid-electric vehicles have lower damages than conventional gasoline/diesel vehicles (baseline technologies) across all counties. This is because hybridelectric vehicles operate more energy-efficiently, thus consuming fewer fuels and leading to lower energy use and emissions from the upstream activities. CNG vehicles or battery-electric vehicles do not always lead to damage reductions. With the 2014 electricity grid, battery-electric vehicle is the preferred technology in some counties while the least-preferred technology in other counties for cars and SUVs ( Figure S4 in the supplementary document). However, if coal power plants are entirely phased out ('alternative electricity grid'), the distributions of damages for battery-electric vehicles would move entirely to the left of those for the other technologies. In other words, battery-electric vehicles powered by coal-phased-out electricity grid would deliver substantial health and climate benefits across the United States. Furthermore, as natural gas compressors, used in the transportation and storage of natural gas as well as compression of natural gas to make CNG fuel, consume electricity for operation, the life cycle damages of CNG vehicles would also reduce under the 'alternative electricity grid.' The frequency distributions of life cycle damages of any specific fuel-vehicle technology are similar for passenger cars and for SUVs. But this is not the case for transit buses. First, CNG transit buses are more likely to increase damages compared to conventional diesel buses due to fuel efficiency penalty and methane leakage along the natural gas supply chain (Tong et al 2015b(Tong et al , 2017. Second, battery-electric buses are more likely to reduce damages compared to conventional diesel buses thanks to stop-and-go duty cycles (that are more beneficial to battery-electric vehicles) and higher tailpipe emissions of criteria air pollutants for conventional diesel buses (Tong et al 2015b(Tong et al , 2017.

'Best' fuel-vehicle technology for each county
To facilitate decision-making, we compare and rank the fuel-vehicle technologies by their life cycle damages for the same vehicle type in the same county. In figure 2, we show the preferred fuel-vehicle technologies that lead to the lowest damages in any county. In particular, we consider three environmental goals: climate change only, air pollution only, and climate change and air pollution. We find that battery-electric buses are the preferred technology across the country for the climate change goal. However, for other goals and other vehicle types, preferred fuel-vehicle technologies depend on the environmental goal considered as well as where vehicles are used.
Looking across environmental goals (i.e. different columns in figure 2), for passenger cars, SUVs, or transit buses, there is no fuel-vehicle technology that achieves the lowest climate change damages and the smallest air pollution damages in all counties. Instead, we identify a trade-off for technology solutions that mitigate climate change damages and those that reduce air pollution damages. Although alternative fuel and advanced vehicles almost always reduce climate change damages compared to conventional gasoline/diesel vehicles, they can increase air pollution damages in some regions (see supplementary document, section 4.4, and Supplementary Data File for quantitative details). It is worth noting, however, that battery-electric passenger car could lead to cobenefits (mitigating both climate change and air pollution damages) for the west coast, Rocky Mountain, and New England. Similarly, gasoline/diesel hybridelectric cars and SUVs would provide co-benefits in the Midwest.
Vehicle electrification has great potential to reduce climate change damages. Battery-electric vehicles lead to the lowest climate change damages for passenger cars and transit buses in the majority of U.S. regions (except the Midwest). Batteryelectric SUVs have lower climate change damages than gasoline and CNG SUVs in all areas. However, significant variability and uncertainty exist in the air pollution damages caused by battery-electric vehicles across the U.S. For regions where clean power plants are already in place, such as the west coast and Rocky Mountain region, battery-electric passenger cars achieve lower air pollution damages than other fuel-vehicle technologies. It is also worth noting that battery-electric SUVs and battery-electric transit buses only lead to the smallest air pollution damages for densely-populated regions on the west coast. The differences in the relative air pollution damages of battery-electric vehicles across vehicle types suggest the need to consider vehicle type in the policymaking.
The emissions profiles of natural gas vehicles are different from conventional gasoline/diesel vehicles. Although natural gas vehicles have lower tailpipe CO 2 emissions and NO x emissions than gasoline/diesel vehicles, they have more substantial life cycle methane emissions (a potent greenhouse gas) and much higher tailpipe CO emissions (a criteria air pollutant). As a result, when comparing life cycle damages of CNG vehicles with gasoline/diesel vehicles, the spatial variability in marginal air pollution damages (primarily determined by baseline emissions, wind speed and direction, and population density) plays an important role. In this study, CNG vehicles always lead to greater damages than gasoline/diesel hybrid-electric vehicles for climate change damage or air pollution damage.

Sensitivity to time of charging
In contrary to petroleum and natural gas fuels, the environmental impact of battery-electric vehicles is much more sensitive to time of vehicle charging. This is because the cost-minimized dispatch of electric power grid varies from hour to hour due to changes in demand, generator availability, renewable energy generation, and transmission line congestion (Siler-Evans et al 2012). The preceding discussion so far highlights the potential of battery-electric vehicles in reducing climate change and air pollution damages, assuming the average emissions from all electricity generation sources in 2014. The concept of 'average electricity' may apply if the adoption of batteryelectric vehicles leads to a sizable electricity load comparable to existing electricity loads. For a limited penetration of battery-electric vehicles, the expected charging load is more likely to cause a 'consequential' change (or perturbation) in the grid operation. To quantify these 'consequential' impacts, we estimate the 'consequential' generation at the margin following the regression-based approach proposed in (Siler-Evans et al 2012). In figure 3, we show that the 'consequential' life cycle climate change and air pollution damages of battery-electric vehicles vary substantially depending on when vehicle charging happens during a day. In Dallas, for instance, the life cycle air pollution damages of battery-electric cars would vary by three times between charging at 4 am and 11 am. In any county within a given electricity grid region, the relationship between the consequential impacts and hour of charging is similar because we assume electrons flow freely within each grid region. However, we find inconsistent patterns regarding the effect of charging time across electricity grid regions ( figure  3). This finding further highlights the need to follow a regionalized strategy to incentivize vehicle charging at certain times to maximize social benefits. Finally, to bound the impact of charging time, we estimated the range of 'consequential' life cycle damages of batteryelectric vehicles when charged at the 'cleanest' and the 'dirtiest' hours in each electricity grid region across the country in the supplementary document (sections 3.3 and 4.2). . While the social cost of carbon would be impacted by the assumed value of a statistical life, we assume, in the following parametric analysis, that they are independent of each other.

Sensitivity to monetary valuation
In figure 4, we show fraction of the U.S. population with the preferred fuel-vehicle technologies (i.e. achieving the lowest climate change and air pollution damages) in their home counties. For instance, for passenger cars (the first column), with the baseline assumptions of the value of statistical life and social cost of carbon (the dashed vertical line), 30% of U.S. population would find batteryelectric vehicles to have the lowest life cycle damages in their home county whereas the other 70% of residents would prefer gasoline hybrid-electric vehicles for the same goal. This result relies on the same underlying results as those shown in figures 1 and 2 but is presented in a compressed way to allow for a visual comparison of results over any changed assumption.
For passenger cars and SUVs, battery-electric vehicles and gasoline-hybrid vehicles are preferred technologies, and their relative shares as the preferred technology for U.S. populations are largely stable across the assumed range of either social cost of carbon or value of a statistical life. However, for transit buses, the preferred technology choices change dramatically depending on the social cost of carbon and the value of a statistical life used.
As we assume the climate change damages are proportional to the social cost of carbon and the air pollution damages proportional to the value of a statistical life, different assumptions for these two variables represent varying weights between climate change damages and air pollution damages in decision-making. In the extreme case, a $0/metric ton social cost of carbon indicates that only air pollution damages are considered, and a $0 value of a statistical life suggests that only climate change damages matter. Since different fuel-vehicle technologies are preferred for transit buses under different goals (figure 2), divergent subjective judgments of decision-makers only amplify this trade-off in decision-making.

Sensitivity to air pollution damage model
There is a considerable difference in the estimated life cycle air pollution damages using the EASIUR model or the AP2 model (Heo et al 2016a, Gilmore et al 2019. In particular, the two models differ by multiple times in their respective valuations of NO x and SO x across the United States due to different baseline emissions inventories and population data, and distinct approaches in calculating the resulting airborne concentrations of PM 2.5 and ozone in the response of criteria air pollutant emissions (Heo et al 2016a, Gilmore et al 2019. For the scope of this paper, the implications of inconsistent estimates from these two models include different damage estimates for the same fuel-vehicle technologies, and, as a result, different rankings of fuel-vehicle technologies in some counties.

Sensitivity to emissions inventory
As summarized in the supplementary document (table S1), all existing studies used the GREET model for emissions inventories. In the supplementary document (section 4.4), we estimated the life cycle air pollution and climate change damages using the emissions data from the GREET model. The GREET model reported higher criteria air pollutant emissions from oil refining than the National Emissions Inventory, so using the GREET model's emissions data would favor alternative fuels and advanced vehicle technologies.
The literature reported different cradle-to-grave air pollution damages for battery manufacturing (by a factor of three) (Michalek et al 2011, Tessum et al 2014. Using the higher battery manufacturing damage estimate leads to 13%-104% increases in the life cycle air pollution damages for battery-electric vehicles across vehicle types and counties compared  (Azevedo et al 2020), and the EASIUR model.
to those using the lower battery manufacturing damage estimate (section 4.3 in the supplementary document). Assuming more serious battery manufacturing damages, the trade-offs between climate change damages and air pollution damages become more evident, but battery-electric cars and buses used in regions with clean electricity still provide overall environmental benefits.

Conclusions and discussion
This paper presents a systematic, spatially-explicit assessment of environmental externalities caused by alternative fuels and advanced vehicle technologies in the United States. Using a coupled modeling framework linking state-of-the-art life cycle assessment, reduced-form air quality models, and integrated assessment models, we estimate the monetized damages in climate change and air pollution caused by the life cycle air emissions from passenger-moving vehicles (passenger cars, SUVs and transit buses). We find substantial spatial variability in the monetized damages for all fuel-vehicle technologies studied.
None of the fuel-vehicle technologies leads to the lowest climate change damages and the smallest air pollution damages across all U.S. counties. Tradeoffs between air pollution mitigation and decarbonization for the fuel-switching strategy (replacement of incumbent petroleum fuels with alternative fuels or advanced vehicle technologies) are persistent across several spatial scales (counties, states, and electricity grid regions). These findings suggest that policy actions towards sustainable transportation should account for the spatial heterogeneity in climate change and air pollution impacts as well as reflect stakeholders' value judgment in terms of the relative importance between clean air and decarbonization goals. As a result, depending on the decisionmaking goal (clean air or stabilizing climate), the preferred technology choices are sensitive to economic assumptions such as the social cost of carbon and the value of a statistical life. We note that such trade-offs could be high hurdles for the adoption of and support for sustainable transportation technologies. Further research should study how these trade-offs impact consumer behaviors and real-world decision making.
Vehicle electrification has substantial potential to reduce climate change damages and air pollution damages. With the 2014 electricity grid, vehicle electrification can already mitigate climate change damages compared to conventional petroleum vehicles on the west coast and New England. However, batteryelectric vehicles can lead to up to three point six times increases in air pollution damages for passenger cars and SUVs and up to six point three times increases for transit buses compared to conventional gasoline/diesel vehicles in U.S. regions with high shares of coal-fired power plants (such as Midwest and Southeast). Even in U.S. regions with a relatively clean electricity grid (such as the west coast and New England), battery-electric vehicles can only partially reduce air pollution damages. These findings highlight the importance of continually cleaning and decarbonizing electricity grid, such as with increased penetrations of renewable energy technologies and nuclear power (Wei et al 2013). A clean electricity grid with near-zero emissions not only benefits the electricity sector and traditional electricity consumers such as buildings but also becomes increasingly crucial for a sustainable transportation future (Wei et al 2013, McCollum et al 2014.
This work highlights the importance of considering the interactions between electricity grid operation and electric vehicle adoption and usage. However, given the limitation in time and scope, we were unable to model explicitly how vehicle electrification would impact the operation of the grid in the near term or how vehicle electrification would lead to new generation assets and transmission lines over an extended time. As the adoption of battery-electric vehicles grows, the charging load from vehicle electrification would start to emerge beyond the 'marginal' electricity load in certain regions and specific hours. As a result, the optimal dispatch of the electricity grid would change, and the new generation capacity would have to be built. Thus, if well planned, vehicle electrification could present new opportunities to decarbonize the electricity grid and the transportation sectors at the same time. Further research could shed light on how to plan and safeguard a sustainable transition of the electricity grid and vehicle fleets. Furthermore, future research should explore technology deployment, policy support, and infrastructure lock-in in the interconnected electricity and transportation systems.
Our work shows the effectiveness and importance of vehicle fuel efficiency and tailpipe emissions standards (Shindell et al 2011). Hybrid-electric vehicles provide tangible benefits in terms of reduced climate change damages and air pollution damages no matter where they are driven. They are also the preferred technologies for regions with relatively dirty electricity grid. More importantly, the increasing performance of energy-efficient vehicles would set the bar with which advanced and emerging vehicle technologies compete. Reducing vehicle tailpipe emissions has long been a strategy to solve the air quality issues in the U.S. (US EPA 2001, U.S. Environmental Protection Agency (EPA) and U.S. Department of Transportation (DOT) National Highway Traffic Safety Administration (NHTSA) 2012). Except for battery-electric vehicles, which eliminate tailpipe emissions, tailpipe emissions still account for a majority share in life cycle emissions inventory and the resulting life cycle monetized damages. Strengthening vehicle tailpipe emissions can be an effective means to reduce vehicles' environmental externalities as vehicles last for more than a decade on average.
Systems analysis that considers multiple different impacts across the life cycle of fuel-vehicle technologies would not be possible without the recent progress in the quantification of health and climate change damages. However, any air quality and impact model would still face the daunting challenge to balance accuracy, representation (e.g. temporal and spatial resolution and scope, atmospheric physics, and chemistry), and computational efficiency. Fullscale chemical transport models (such as CMAQ) strive to simulate the best atmospheric chemistry and quantify the most detailed spatial and temporal representations at the expense of high computational resources. As a result, the use of such models in our study would limit sensitivity analysis and fail to identify the most sensitive factors or variables. Reduced-form air quality and impact models, such as those we have used in this paper, significantly reduce the running time, but at the cost of coarser spatial representation and approximation of the atmospheric chemistry. In particular, the countylevel spatial resolution of the EASIUR and AP2 models would underestimate the health impacts of tailpipe emissions to drivers, pedestrians, and residents who are near road infrastructure. The lack of finer-scale representation would bias results in favor of gasoline/diesel vehicles versus battery-electric vehicles whose emissions occur at remote locations of power plants.
Finally, we note that a comprehensive analysis to determine fuel-vehicle technologies for passengercarrying vehicles should include other fuel technologies (e.g. biofuels, hydrogen, and fuel cell electric vehicles) and consider different perspectives such as economics, consumer behavior, vehicle use patterns, and infrastructure planning (Wolinetz et al 2018, Tong et al 2019. Nevertheless, this work shows that the life cycle climate change and air pollution damages of fuel-vehicle technologies can and should be considered. Quantifying and internalizing externalities caused by greenhouse gas and air pollutant emissions from the life cycle of vehicles is essential to achieve sustainable transportation.

Acknowledgments
We are very grateful to Paulina Jaramillo's insights and input to this work, which helped frame and organize much of this paper. We thank Peter Tschofen for sharing the source code for processing the raw data of the National Emissions Inventory and the source classification code database. We thank Steve Davis for insightful discussions on how to present the results. We would like to acknowledge the 2013-14 Northrop Grumman Fellowship, the 2013-14 Steinbrenner Institute Graduate Research Fellowship, the 2016 Ji Dian Liang Fellowship, the Fuels Institute, a gift funding from Toyota Corp., the Center for Climate and Energy Decision Making (CEDM, a cooperative agreement between the National Science Foundation (SES-0949710) and Carnegie Mellon University), and the Center for Air, Climate and Energy Solutions (CACES, a center established under Assistance Agreement No. R835873 awarded by the U.S. Environmental Protection Agency). This work has not been reviewed by EPA or any other funding organization. This work is solely the responsibility of the authors and does not reflect any of the funding organizations. EPA does not endorse any products or commercial services mentioned in this publication.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.