Potential impacts of electric vehicles on air quality and health endpoints in the Greater Houston Area in 2040

Significant emissions from transportation contribute to the formation of O 3 and fine particulate matter (PM 2.5 ), causing poor air quality and health. In this study, four scenarios were developed to understand how future fleet electrification and turnover of both gasoline and diesel vehicles affect air quality and health in the Houston Metropolitan area. These scenarios considered increased vehicle activity and various configurations of emissions controls. Comparing to a base year of 2013, model predictions for 2040 indicated a ~50% emissions increase in the Business As Usual (BAU) case, and ~50%, ~75%, and ~95% reductions in the three distinct emissions control cases, the Moderate Electrification (ME), Aggressive Electrification (AE), and Complete Turnover (CT) cases, respectively. Each modeling scenario was conducted using a high-resolution (1 km) WRF-SMOKE-CMAQ-BenMAP air quality and health modeling framework, which helped capture urban features in higher detail. The emissions control cases resulted in 1-4 ppb maximum 8h O 3 increase along highways and reductions both in the regions enclosed by the highways and those downwind. Simulated PM 2.5 concentrations decreased between 0.5-2 μg m -3 . Health impact results suggest that increased O 3 and PM 2.5 concentrations from the BAU case will lead to 122 additional premature deaths with respect to 2013. However, reduced emissions for the control cases (ME, AE, CT) will prevent 114-246 premature deaths. Additionally, about 7,500 asthma exacerbation and 5,500 school loss days will be prevented in the ME case, benefiting younger individuals. The economic benefits generally followed the same trends as health impacts. The analysis framework developed in this study can be applied to other metropolitan areas. The effects of motor vehicle electrification on power plant emissions were estimated using the Argonne National Laboratory’s Autonomie data, and indicated the electrification load to be negligible as opposed to projected electricity generation. Abstract Significant emissions from transportation contribute to the formation of O 3 and fine particulate matter (PM 2.5 ), causing poor air quality and health. In this study, four scenarios were developed to understand how future fleet electrification and turnover of both gasoline and diesel vehicles affect air quality and health in the Houston Metropolitan area. These scenarios considered increased vehicle activity and various configurations of emissions controls. Comparing to a base year of 2013, model predictions for 2040 indicated a ~50% emissions increase in the Business As Usual (BAU) case, and ~50%, ~75%, and ~95% reductions in the three distinct emissions control cases, the Moderate Electrification (ME), Aggressive Electrification (AE), and Complete Turnover (CT) cases, respectively. Each modeling scenario was conducted using a high-resolution (1 km) WRF-SMOKE-CMAQ-BenMAP air quality and health modeling framework, which helped capture urban features in higher detail. The emissions control cases resulted in 1-4 ppb maximum 8h O 3 increase along highways and reductions both in the regions enclosed by the highways and those downwind. Simulated PM 2.5 concentrations decreased between 0.5-2 μg m -3 . Health impact results suggest that increased O 3 and PM 2.5 concentrations from the BAU case will lead to 122 additional premature deaths with respect to 2013. However, reduced emissions for the control cases (ME, AE, CT) will prevent 114-246 premature deaths. Additionally, about 7,500 asthma exacerbation and 5,500 school loss days will be prevented in the ME case, benefiting younger individuals. The economic benefits generally followed the same trends as health impacts. The analysis framework developed in this study can be applied to other metropolitan areas. The effects of motor vehicle electrification on power plant emissions were estimated using the Argonne National Laboratory’s Autonomie data, and indicated the electrification load to be negligible as opposed to projected electricity generation.


Introduction
Estimating transportation emissions trends and associated air quality impacts can provide important insights for requisite control policy. The transportation sector is a major contributor to the concentrations of both nitrogen oxides (NOx) and volatile organic compounds (VOCs), which react in the presence of sunlight to form ozone (O3). Vehicular traffic also emits important components of fine particulate matter (PM2.5) such as organic and elemental carbon (Roy et al., 2011;U.S. EPA, 2017a). Both O3 and PM2.5 are known to be harmful to human health, causing premature deaths (Bell et al., 2005;Krewski et al., 2009) and severe and minor morbidities (e.g., hospital admissions and asthma exacerbations) (Katsouyanni et al., 2009;Mortimer et al., 2002). The adoption of electric vehicles can lead to significant emissions reductions in the transportation sector (Huo et al., 2012;Nichols et al., 2015). As of 2017, electric vehicles accounted for ~1% of new car sales in the United States (and much higher in other markets, such as Norway with nearly 30% market share) (Vaughan, 2017). A Bloomberg New Energy Finance report (BNEF, 2016) estimated that 35% of global new car sales would be electricpowered in 2040, that was adjusted to 54% in 2017 considering falling battery costs and possible consumer adoption growth (BNEF, 2017). Policymakers from several countries have announced their commitments of a transition to electric vehicles. For example, Norway has called for all new car sales to be electric by 2025; France, the U.K., and the state of California set goals to achieve the same by 2040; and China aims to electrify 20% of its new cars by 2025 (Walsh, 2017). The implementation of these policies will have large impact on transportation emissions, air quality, and human health.
Several previous studies have assessed the potential impacts of electric vehicles on air quality in various regions. Thompson et al. (2009) found that replacing 20% of the gasoline vehicles with electric vehicles over the northeastern U.S. could reduce MDA8 O3 by 2-6 ppb, barring a few episodes of significant O3 increases around Newark and Philadelphia. Brinkman et al. (2010) reported that 100% penetration of electric vehicles in Denver would lower the summertime MDA8 O3 by 2-3 ppb, excluding a 1-2 ppb increase over the downtown area.  reported that fully electrifying all light-duty vehicles would lead to reductions in MDA8 O3 of 1-2 ppb and mean peak-time O3 of as much as 7 ppb across Taiwan, except in central metropolitan Taipei (an increase of < 2 ppb). Their findings also indicated that electrification would reduce the annual number of days of O3 pollution episodes by 40% and PM2.5 pollution episodes by 6-10%. Nopmongcol et al. (2017) indicated that 17% of light-duty and 8% of heavy-duty vehicle miles traveled and 17-79% of various off-road equipment types would be targeted for electrification in the U.S. by the year 2030. They reported that electrification would reduce O3 by less than 1 ppb and PM2.5 by 0.5 µg m -3 .
To date, no similar quantitative evaluation has been conducted for the Houston Metropolitan Area. This is the 4 th largest urban area in the United States and classified as a nonattainment area for ozone (O3) by the United States Environmental Protection Agency (U.S. EPA) (U.S. EPA Green Book, 2017). It is classified as "Marginal" with a current design value (2015-2017) of 80 ppb. It is also in borderline attainment for fine particulate matter (PM2.5). As indicated by the 2013 Houston Galveston Area Council (H-GAC) Regional Goods Movement Plan, the population of the Houston area is projected to grow by 50% in 2040, which will result in increased motor vehicle activity and goods movement in the region if the projections hold true. Their study also projected a doubling of the number of trucks in the region by 2040. Another study by the Texas Transportation Institute projected that the number of trucks in the Houston area would increase by 40%-80% and the number of gasoline vehicles by 30% to 50% by 2040 (TCEQ, 2015). Thus, the Houston Metropolitan Area provides a valuable opportunity to evaluate the impact of increased transportation activity and control strategies (e.g., fleet electrification) on air quality and the subsequent health effects.
In this study, we apply relatively large fleet penetrations of electric vehicles, including a moderate fraction (35%) and an aggressive one (70%), to on-road mobile source sectors in the Greater Houston Area for the year 2040. For assessing the future emissions, another important factor that needs to be considered is fleet turnover, where older motor vehicles are replaced with newer technology, resulting in a significant reduction in emissions (Roy et al., 2014;Liu et al., 2015). Previous studies have assessed the impact of electrification on transportation emissions (Huo et al., 2012;Nichols et al., 2015) and air quality (Thompson et al., 2011;Soret et al., 2014;Nopmongcol et al., 2017), but few have investigated the corresponding health and economic benefits. Pan et al. (2017b) developed and evaluated a fine-resolution (1×1 km) WRF-SMOKE-CMAQ air quality modeling system to gain an understanding of the concentrations of O3 and the processes over Houston that drove O3 formation in September 2013. In this study, we extend this framework by adding a health/benefits model to assess motor vehicle emissions, fleet electrification and turnover, and their associated air quality and health impacts. This study aims to answer the following questions: (1) How would the mobile activity (e.g., vehicle miles traveled) change in the Houston area by 2040? What are the possible future scenarios corresponding to varying degrees of emission control, fleet electrification and turnover for gasoline and diesel vehicles? Under these scenarios, how would the emissions change in 2040?
(2) What are the magnitudes and spatial distributions of changes in regional O3 and its precursors as well as PM2.5 and its speciated components?
(3) What are the magnitudes of changes in premature mortality, morbidities, and consequent economic benefits? What are the primary factors affecting the health impact results? How do the health results vary by county?

Air quality modeling system setup and observational data
This study employed a WRF-SMOKE-CMAQ modeling system that included meteorological, emissions, and chemical transport modeling (a simplified flow chart in Fig. S1). Here, we used the U.S. EPA Community Multi-scale Air Quality (CMAQ) model (Byun and Schere, 2006) version 5.0.2 to model chemical transport, the Carbon Bond 5 (CB05) (Yarwood et al., 2005) and AERO6 mechanisms to simulate gas and aerosol chemistry, respectively, and the Weather Research and Forecasting (WRF) model version 3.7 (Skamarock and Klemp, 2008) to simulate meteorology. This study used the National Centers for Environmental Prediction (NCEP) North American Regional Reanalysis (NARR) data (Mesinger et al., 2004) as input for the WRF model. Emission inputs were prepared using the Sparse Matrix Operator Kernel Emissions (SMOKE) system version 3.6 (Houyoux et al., 2000) using the U.S. EPA 2011 National Emission Inventory (NEI-2011) (U.S. EPA, 2015a; 2015b). Details of WRF physics options and CMAQ configurations are listed in Pan et al. (2017b) and used in previous air quality modeling studies Pan et al., 2017a).
The 1 km resolution simulation domain covered the eight-county Houston-Galveston-Brazoria (HGB) nonattainment area depicted in Fig. 1. Initial and boundary conditions were obtained from the Air Quality Forecasting system at the University of Houston (AQF-UH) (http://spock.geosc.uh.edu/, Choi et al., 2016) over a larger 12 km grid covering the United States, southern Canada, and northern Mexico (left panel in Fig. 1). Model-measurement evaluation was conducted using the Texas Commission on Environmental Quality (TCEQ) Continuous Ambient Monitoring Stations (CAMS) network, which provided in-situ ground-level measurements for several chemical species such as NOx, O3, and a large suite of VOCs, and the U.S. EPA AirNow network, which provided surface PM2.5 measurements.

Fig. 1.
Horizontal domains of WRF and CMAQ at various grid resolutions; the HGB 1 km is used in this study while the U.S. 12 km is used to provide boundary conditions. In the zoomed-in plot on the right, roadways are represented in orange and county boundaries in purple.

Mobile source emissions processing
The air quality simulations in this study used emissions from all major sectors, including mobile, point, area, and biogenic sectors. Here, the calculations of on-road mobile emissions were specifically described as we focused on the future change in the transportation sector. This study employed the U.S. EPA Motor Vehicle Emissions Simulator (MOVES) (U.S. EPA, 2016) and SMOKE models to prepare on-road mobile emissions. The MOVES model was ran by U.S. EPA to provide emission factors (EF) look-up tables for mobile emissions as a function of speed, fuel type, vehicle type, road type, and meteorological conditions. The major fuel types include gasoline and diesel, and the vehicle types include motorcycle, passenger car/truck, light commercial truck, intercity/transit/school bus, and medium/heavy duty short-haul/long-haul truck.
In Fig. 2, the "rate-per-distance" (RPD) sector represents on-roadway mobile sources and emission processes. These emissions are located along the roadway networks inside the simulation domain. RPD emissions can be calculated using the following simplified formula: where the unit of is g mile -1 and represents vehicle miles traveled.
The "rate-per-vehicle" (RPV) sector represents off-network mobile emissions, which were predominantly concentrated in the urban locations and estimated as follows: where the unit of is g vehicle -1 hour -1 , and denotes the vehicle population in numbers.
Two other mobile sectors were included: the "rate-per-profile" (RPP), which represents offnetwork vapor-venting emissions from parked vehicles, was estimated in the same way as RPV; and the "rate-per-hour" (RPH), which includes only extended idle exhaust emissions from on-roadway long-haul trucks, also known as "hoteling". The unit of the emission factor for the RPH is g hour -1 , and the RPH's activity data are hoteling hours (Hoteling). The activity data (e.g., VMT, VPOP, and Hoteling) came from NEI-2011 and varied by county, month, and source attributes such as fuel type, vehicle type, road type, and emission processes. The activity data were originally submitted by state/local/tribal air agencies to the U.S. EPA. The SMOKE model took EF (the output from MOVES), county-level monthly activity data, and temperature data and produced hourly gridded CMAQ-ready emissions.

Projections of mobile activity data
Future projections for VMT and VPOP were estimated from calculations performed by the Texas Transportation Institute (TTI) for the Texas Commission on Environmental Quality (TCEQ, 2015). The results are plotted in Fig. 3. Here, panel (a) represents the aggregate VMT (gasoline+diesel) for each of the counties, and panel (b) represents the scaling factors, indicating an increase in VMT from 2013 to 2040 by 30%-80% over the eight-county area. Two surrogate gasoline-diesel split profiles, Houston and Beaumont, were developed for the eight-county area. In panel (c), the gasoline-diesel split for VMT was 93%-7% for Houston and 82%-18% for Beaumont in 2013. The split changes marginally in favor of diesel in 2040, 92%-8% for Houston and 81%-19% for Beaumont. The diesel fractions are higher over suburban Beaumont while gasoline activities are significantly higher in the urban region (i.e., the city of Houston). The Houston profile was used to represent the Brazoria, Fort Bend, Galveston, Harris, Montgomery, and Waller Counties and the Beaumont profile was used to represent the Chambers and Liberty Counties. The projected gasoline and diesel VMT in 2040 are plotted in panels (d) and (e), with their specific scaling factors in (f). The projected gasoline VMTs are roughly one order of magnitude higher than diesel VMTs because of the greater population of gasoline-powered vehicles. The projected gasoline and diesel scaling factors closely mirror the total VMT, indicating that the change in VMT is more significant than that in the gasoline-diesel split. One subtle difference: the diesel scaling factors are slightly magnified, while the gasoline ones are slightly depressed. For example, in Harris County, the total VMT changes by a factor of 1.46, while the diesel VMT changes by 1.59 and gasoline by 1.45. This slight magnification in diesel scaling factors could be attributed to the marginal shift in favor of diesel (~9% increase) as shown in panel (c). The projected VMT profiles were also used for county and fuel-specific vehicle population (VPOP) and hoteling projections.

Design of scenarios
This study designed several emissions scenarios to account for uncertainties in fleet turnover and electrification. In Table 1, "New" indicates the percentage of the fleet in 2040 that uses or is retrofitted with state-of-the-art combustion and emission control technologies. "Electric" represents the percentage of fleet comprising electric vehicles, and "Current" represents the fraction carrying over from the base year of 2013 that is neither retrofitted nor replaced. The "Scaling factor" in the last column is a function of both control technology efficiency and fleet turnover, applied to aggregate (distance, vehicle, profile, and hoteling) gasoline and diesel emissions. The simulation scenarios are as follows: (1) A "Base year" (BASE) case, representing the month of September 2013, serves as the base case to which future scenarios were compared. (2) A "Business As Usual" (BAU) case, representing the "worst case" scenario, in which all vehicles in 2040 will have the same emission factors as in 2013. (3) A "Moderate Electrification" (ME) case, representing a mix where 35% of vehicles will be electricpowered based on the assumptions of a Bloomberg New Energy Finance report (BNEF, 2016), and the remaining fleet will be equally apportioned between vehicles with emissions factors like those in 2013 and new vehicles with improved emission factors. (4) An "Aggressive Electrification" (AE) case, representing a mix in which 70% of the Houston region fleet is electrified, and the remaining fleet are equally apportioned between vehicles with emissions factors like those in 2013 and new vehicles with improved emissions factors.
(5) A "Complete Turnover" (CT) case, representing the "best case" scenario, in which the total fleet comprise either state-of-the-art combustion technology or electric vehicles. For all future year scenarios, mobile activity data (e.g., VMT, VPOP, and Hoteling) were increased to the 2040 level, and projections of coal and natural gas EGUs were also applied.

Scaling factors for mobile emission factors
The effective emission factor calculation as described in the U.S. EPA's National Mobile Inventory Model (NMIM) is as follows: where (2040) and (2013) are the emission factors for 2040 and 2013, respectively, represents the cumulative fraction of the fleet that has been replaced by newer, lower emitting sources between 2013 and 2040, and represents the fractional reduction of emissions brought about by this fleet replacement. The value of was assumed to be 95% based on Roy et al. (2014). To include the portion of electric vehicles, the original Equation (3) was modified as follows: where , , and are the percentage fleet turnover values in Table 1. The scaling factor values listed in Table 1 were obtained using equation (4). For instance, the scaling factor in the ME case was:0.33 × (1 − 0.95) + 0.35 × 0 + 0.32 = 0.3365. It should be noted here we assumed zero emissions from electric vehicles for easier demonstration. Electric vehicles produce no tailpipe exhaust emissions. However, non-exhaust emissions (e.g., particulate matter emissions from brake and tire wear) would not be significantly reduced and require technology improvement from those emission modes (Soret et al., 2014).

Electricity demand and projections of EGUs
The added electricity required to power the motor vehicle fleet could potentially result in increased emissions from Electricity Generating Units (EGUs). However, projections by the Electricity Reliability Council of Texas (ERCOT, Borkar, et al., 2016) have indicated that the projected electricity generation will be in western Texas, resulting in no new emissions in the eight-county area. Additionally, ERCOT projections indicate a significant retirement of fossil-fired capacity for southeastern Texas. Despite adding future capacity in the Houston region in our simulations, we needed to account for capacity downsizing in order to represent a more realistic scenario in 2040. Assuming a linear decline trend starting from 2013, we estimated factors of 0.89 and 0.99 (11% and 1% reductions) and applied them to future EGUs powered by coal and natural gas, respectively.
Even though ERCOT projections indicate no new emissions in the Houston area, it is still necessary to calculate the incremental electricity demand due to the vehicle electrification. Here, the calculation was performed for the Aggressive Electrification scenario, which represented the highest electricity consumption for electric vehicles and hence provided bounding estimates. The VMT data from Fig. 3 were fractionated into New, Electric, and Current types. Each vehicle type was multiplied using its specific energy consumption value per unit mile. Projections for electricity consumption (Wh mile -1 ) for cars were taken from the Argonne National Laboratory (ANL)'s Autonomie model output. For this, the Autonomie data on vehicle types had to be mapped to its MOVES counterpart. For example, the energy consumption for a MOVES passenger car was assumed to be half that of Autonomie compact and midsize combined. The mapping is detailed in Table S1. Electricity data for trucks were adopted from the ANL's AFLEET model (Burnham, 2017). The energy consumption values (in Table S2 and S3) for each vehicle type, per unit mile, were multiplied with their specific VMT values and summed up across all vehicle types to get a total electricity requirement. Additional details for this calculation are listed in Section S2 of the Supplemental Information. The calculations for additional electricity requirement indicated projected electrification load to be negligible (< 1%) of the existing power plant generation. The reasons behind the low fraction could be the fact that Autonomie projections account for improved technology and efficiency in 2040 representing lower energy consumption. This coincides with the BNEF calculations that even as EVs represent 53% of global new car sales, they only take up a small faction (~5%) of global power generation in 2040 (BNEF, 2017). Hence, the incremental electricity demand was not considered in our simulations.

The health impact model
To evaluate the impact of these various scenarios on health and economic costs in the year 2040, we used the U.S. EPA Environmental Benefits Mapping and Analysis Program (BenMAP) Community Edition version 1.3 (U.S. EPA, 2017b). The air quality inputs of the model include a baseline scenario (without control) and a control scenario (with an emission control policy implemented). In this study, the base year case (2013) is the baseline scenario and the future year cases are different control scenarios. The health impact calculations in BenMAP are based on concentration-response (C-R) functions, typically representing a decrease in adverse health effects with a concentration of air pollutants (Fann et al., 2012). One group of widely used C-R functions are in the log-linear format: where ∆ represents the change in the incidence of adverse health effects, the concentration-response coefficient, ∆ change in air quality (e.g., O3 concentrations), 0 the baseline incidence rates, and the affected population. The relationship between changes in air pollutant concentrations and incidence of health outcome (i.e., ) are usually assessed in epidemiological studies. Additionally, the BenMAP model calculates the economic cost of avoided premature mortality using a "value of statistical life" (VSL) approach, which is the aggregate monetary value that a large group of people would be willing to pay to slightly reduce the risk of premature death in the population (U.S. EPA, 2017b). The economic costs for morbidities were estimated using the cost of illness, which includes direct medical costs and lost earnings associated with illness.
In our analyses, two pollutant metrics were used: maximum daily 8-hr average (MDA8) for O3 and daily 24-hr mean for PM2.5. We used the PopGrid program (U.S. EPA, 2017b) to allocate the 2010 block-level U.S. Census population data to our study domain. The county-level population growth rates for each year from 2000 through 2050 were developed by Woods & Poole (2015) and pre-installed in the BenMAP model. We first evaluated the health endpoint of "Mortality, All Cause". For O3, we chose C-R coefficient based on the epidemiological studies by Bell et al. (2005), Zanobetti and Schwartz (2008), and Levy et al. (2005), and for PM2.5, we chose a study by Krewski et al. (2009). These studies were chosen as their analyses were based on a large geographic area (e.g., 116 U.S. cities in Krewski et al. (2009)). Hence, they are likely to be more representative and applicable to our analysis in the Houston area. Moreover, we also examined several O3-induced morbidities (e.g., asthma exacerbation, emergency room visits). Because the health impact functions for morbidities were derived from fewer cities or smaller time-scale sample sizes, the functions from several epidemiological studies (listed in Table 2) were used to estimate the risk outcome.

Evaluation of the base year modeling
Domain-wide statistics of model-measurement comparisons for surface O3 and its precursors between base year model output and in-situ measurements for September 2013 are listed in Table 3. The results are similar to the 4 km simulations of  and Pan et al. (2015;. The timeseries comparisons are presented in Fig. S3. For each species, the simulated concentrations generally reproduced the daily variations in the observations during the simulation period.

Changes in on-road mobile emissions in the future scenarios
Because the emissions inventory is "ground-zero" for the modeling framework, a comparison of pollutant emissions for each scenario provides insight into the corresponding air quality changes. Fig. 4 plots the episode-average eight-county aggregated on-road mobile emissions in each simulation scenario for NOx and PM2.5. As expected, the Business As Usual case exhibits significant increases in species emissions with respect to the base year, resulting from an increase in vehicles on the road without emissions controls/retrofits. The other cases show various levels of emission reductions. Table S4 lists base year emissions and percentage changes in future year cases for major pollutants. The VOCs include important air toxic species, such as formaldehyde, acetaldehyde, acrolein, and naphthalene, which have high risk factors for cancer. Compared to the BASE case, the fractional changes are ~50% increase in the BAU case, and ~50%, ~75%, and ~95% reductions in the Moderate Electrification, Aggressive Electrification, and Complete Turnover cases, respectively. These changes in emissions are consistent with the assumptions used to develop these scenarios.

Ozone and its precursors
Because O3 is a secondary pollutant, understanding the distributions of O3 precursors such as NOx and dominant highly reactive VOC species is useful. Fig. S4 lists the descriptions of VOC species in the CB05 chemical mechanism and the averaged VOC speciation from on-road mobile emissions in the BASE case. The major highly reactive VOCs are ethylene (ETH, 4.8%), olefin (OLE+IOLE, 3.5%), toluene (TOL, 2.6%), and xylenes (XYL, 2.8%). The paraffin (PAR, or alkanes) are only weakly reactive with OH and ionic substances, even though they account for ~70% of the total on-road mobile emissions. Fig. 5 plots the spatial distributions of base year surface O3 and its precursors (e.g., NOx, CO, and several highly reactive VOC species). The MDA8 O3 hotspots are located northwest of the urban center, where predominant wind originates from Galveston Bay and the Gulf of Mexico to the southeast (Pan et al., 2017a). High NOx and CO concentrations are located over the urban area because of high motor vehicle emissions. It is important to identify that NOx contours are steeper than CO due to the shorter lifetime of NOx. While elevated ethylene concentrations are predicted over industrial regions, toluene and xylenes concentrations are high over both industrial regions and urban road networks. Formaldehyde hotspots coincide with non-point source oil and gas production regions, and isoprene and terpenes concentrations are high over the northern part of the domain, resulting from significant biogenic emissions generated by large forest cover. Fig. 6 plots the CMAQ-simulated NOx and MDA8 O3 concentrations for each scenario. The results, shown in panels (a)-(d), plot a decrease in absolute NOx concentrations with increasing fleet turnover, electrification, and control. For example, concentration hotspots are predicted over all highway loops over Houston for the BAU case, and are predicted to significantly decrease in the CT case. In other words, partial fleet electrification accompanied by complete fleet turnover result in lower NOx emissions and thus, lower concentrations. Panels (i)-(l), which plot O3 concentrations, tell a different story. The BAU case shows lowered MDA8 O3 concentrations over the highway loops and higher concentrations elsewhere. Because highways have significant NOx emissions, they are NOx-saturated. In such areas, O3 and NOx concentrations are inversely correlated, as previous studies (e.g., Choi et al., 2012) illustrated. In addition, panel (i) illustrates increased O3 concentrations over regions northwest of the loop, resulting from O3 formation in the outflow of NOx-saturated areas. The outflow regions are NOx-limited and provide favorable conditions for O3 formation. With tighter controls, increased fleet turnover, and decreasing NOx concentrations, O3 concentrations increase along the highway loop and decrease over the outflow. Similar findings are corroborated in panels (m)-(p), which show the effects of the O3 impacts vis-à-vis the 2013 base case. It is predicted that O3 concentrations resulting from increased motor vehicle emissions decrease in the BAU case over the NOx-saturated areas by 1-3 ppb while increasing 1-2 ppb over the outflow. Increasing controls, retrofits, electrification, and turnover lead to lower NOx emissions and an increase in O3 concentrations by 1-3 ppb over the highways but decrease over the entire outflow surrounding the highway loop, as well as the areas enclosed by the loop. Notably, the CT case exhibit a decrease of 3-4 ppb over the northwestern outflow, the same region where a significant O3 increase is predicted in the BAU case. Fig. S5 displays the spatial differences for CO, ethylene, and isoprene. The distributions of the change in CO and ethylene concentrations in future year cases are similar to those in NOx concentrations. Whereas large positive changes in simulated isoprene concentrations exist for the control cases over the northern domain, the decrease in isoprene concentrations in the BAU case could be attributed to more indirect reactions with the enhanced NOx emissions and concentrations . In high-NOx conditions, NOx serves as a sink for isoprene. Isoprene reaction with hydroxyl radical (OH) in the presence of nitrogen oxide (NO) forms isoprene nitrate or converts NO to NO2, which can undergo photolysis to form O3.   S6 plots the differences in MDA8 O3 between the CT case and BASE case on selected days with relatively large value differences (~10 ppb over downwind regions). Whereas the positive difference areas are always located over the NOx-saturated urban roadways, the large negative areas vary spatially because of day-to-day changes in wind direction, suggesting daily negative differences (~10 ppb) could be much larger than episode-average MDA8 O3 reductions (3-4 ppb, in the last panel of Fig. 6). Fig. S7 depicts the simulated base year O3 during the peak time (13-16 CST) and morning rush hour (6-9 CST), and Fig. S8, the difference plots, illustrate the corresponding changes in future year scenarios. The spatial distribution patterns of the BASE and difference O3 during the peak time are similar to those of the MDA8 O3, but with higher magnitudes. During the morning rush hour, however, the average O3 concentrations increase by ~10 ppb in the CT case, which results from NOx reductions and less titration (last panel of Fig. S8). The morning O3 increase due to NOx reduction is also reported by the study of vehicle electrification over the Yangtze River Delta region in China by Ke et al. (2017).

PM2.5 and speciated PM2.5
PM2.5 comprises a large number of species which widely differ in properties. Table S5 lists the primary PM2.5 species represented in the SMOKE and CMAQ models. Fig. S9 displays the composition of domain-wide CMAQ simulated ground-level PM2.5 species. The sum of inorganic aerosols (sulfate, nitrate, and ammonium) accounts for 30%, with sulfate being predominant (23%). The sum of primary organic carbon (POC), secondary organic carbon (SOC), elemental carbon (EC), and primary noncarbon mass (APNCOM) constitute 25% of simulated ground-level PM2.5 species. Sodium (Na + ), chloride (Cl -), and trace metals (Fe, Al, Si, Ti, Ca 2+ , Mg 2+ , K + , Mn) occupy 15% in total, with the remaining unspeciated PM2.5 accounting for 30%. Simulated concentrations of PM2.5 and its major speciated components are plotted in Fig. 7 for September 2013. Total PM2.5 concentrations in most regions fall below 20 µg m -3 , except over industrial locations with large clusters of petrochemical and chemical manufacturing facilities. Relatively high concentrations of POC, APNCOM, and EC are predicted over highways, representing significant emissions from both gasoline and diesel motor vehicles, as corroborated by previous studies (e.g. Lane et al., 2007;Roy et al., 2011). SOC includes secondary products from both anthropogenic and biogenic sources (Carlton et al., 2010). Their hotspots are located over forest areas, indicating oxidation of biogenic VOCs (e.g., isoprene and monoterpenes) and aged biogenic aerosols are the major contributors. Elevated sulfate and ammonium concentrations are predicted near industrial sources. Sulfate contributions are likely due to the burning of sulfurcontaining fuels in the Houston Ship Channel and from coal-fired power plants (Zhang et al., 2015). Shimadera et al. (2014) suggested that nitrate can be affected by ammonia (NH3) concentrations and the formation of ammonium nitrate (NH4NO3).  Because of the changes in mobile PM2.5 emissions in the future year cases, understanding the major species that make up mobile emissions is useful and important. The SPECIATE databases (U.S. EPA, 2017a) have reported that elemental carbon (EC) and organic carbon (OC) have high weight fractions for PM2.5 speciation in transportation sources, such as Heavy Duty Diesel Vehicles (HDDV) exhaust (EC: 77%; OC: 17.5%), Light Duty Diesel Vehicles (LDDV) exhaust (EC: 51.4%; OC: 35.5%), and on-road gasoline exhaust (EC: 19%; OC: 55%). Here, we calculated the mass fraction of each PM2.5 component in the aggregated on-road mobile emissions in Table S6. Species with higher fractions will undergo a relatively substantial change in the future year cases, as opposed to those with lower fractions. Similarly, the two biggest components of PM2.5 are elemental carbon (42%) and primary organic carbon (30%). Thus, the changes in total PM2.5 concentrations are mainly associated with changes in carbonaceous species. Fig. 8 plots the differences between the projected scenarios and the base 2013 case. Panels (a)-(d) represent total PM2.5 concentrations. The BAU case results in an increase in PM2.5 concentrations of 1-2 μg m -3 , while the control scenarios show decreases in PM2.5 concentrations ranging from 0.5-2 μg m -3 . The most dramatic changes occur on highways because of reductions in motor vehicle emissions, corroborated in the plots for EC (panels (e)-(h)) and POC (panels (i)-(l)). The changes in sulfate also mirror those of EC and POC, but one additional important point is the reduction in sulfate hotspots over regions with EGU emissions, due to the reduction in coal capacity over these areas. The changes in PM2.5 concentrations could also be attributable to other chemical reactions associated with gas-to-particle conversion (e.g., HNO3(g)→nitrate). In all the control scenarios, the reduced NOx, a precursor for HNO3, contributed to the decreased nitrate concentrations over the downwind northwest (Fig. S10). Moreover, the decreased VOCs emissions in the urban area also lead to reduction in the concentrations of secondary organic carbon.

Evaluation of the changes in health endpoints and benefits
The health impact calculations in BenMAP are based on concentration-response (C-R) functions, which represent a decrease in health incidence with air pollutant concentrations. Based on these calculations, the BAU case would likely result in an increased number of premature deaths with respect to 2013. All of the control scenarios, however, would result in prevented mortality, as shown in Table 4. For PM2.5, the table shows about 121 more premature deaths in the BAU case over greater Houston, and 109, 177, and 229 prevented premature deaths in the ME, AE, and CT cases, respectively. These findings coincide with trends in PM2.5 concentration, as depicted in panels (a)-(d) in Fig. 8. The findings also roughly correspond to a 61% increase in PM2.5 emissions in the BAU case, and 46%, 75%, and 95% reductions in the ME, AE, and CT cases listed in Table S4.
Interpreting the O3 results, however, is more complicated because the trends of O3 change vary spatially (panels (m)-(p) of Fig. 6). For instance, in the BAU case, BenMAP predicts an increase in adverse health effects over downwind areas because of O3 increases, while predicting a decrease of damage in the urban and major highways. The net number of premature deaths is negligible at -0.04, and indicates that the downwind negative impact is barely larger than the positive impacts in the urban area. In contrast, for the other scenarios with emissions reductions (i.e., the ME, AE, and CT cases), the gains in health endpoints in downwind areas are all greater than the losses over the urban highways, resulting in about 5, 11, and 17 prevented premature deaths, respectively. We may expect more health benefits if the simulation domain were to cover more downwind areas. It should be noted that even in the case of an increase in O3 concentrations over the urban highways, reductions in air toxics emissions would occur, which would lead to more health benefits not studied here because the health impact functions for these air toxics are not available in the current BenMAP model. The economic cost (benefit) values generally coincide with premature mortality results.  (2) In the 3 rd and 4 th columns, positive values indicate the number of premature deaths prevented because of control strategies and the associated benefits achieved, while the negative values in the BAU case indicate an increase in the number of premature deaths and economic losses. Fig. 9. The county-level avoided premature mortality (APM) resulting from the changes in O3 and PM2.5 concentrations in the future year scenarios. Fig. 9 plots the county-level avoided premature mortality results. For PM2.5, among all the counties, Harris County (in the middle) exhibits significant larger changes due to its high population and dominant change in PM2.5 concentrations. For O3, the net increase of avoided premature mortality in Harris County in BAU case (2.34) are contributed by O3 reduction over highways, while those in the control cases (1.03, 4.05, and 7.60) result from the O3 reduction over the outflow regions, as corroborated in panels (m)-(p) of Fig. 6. Table 5 shows similar trends in O3-induced morbidities and associated benefits. Thus, the emission reduction scenarios would significantly reduce asthma exacerbation and school loss days, benefiting younger individuals.

Conclusions and discussion
This study modeled multiple scenarios to investigate the effects of control technologies, fleet turnover, and electrification of both gasoline and diesel vehicles on air quality and health over the eightcounty O3 nonattainment region around Houston in the year 2040. Comparing scenarios for the year 2040 to the base year of 2013, we found that emissions increased by ~50% emissions in the BAU case, and they reduced by ~50%, ~75%, and ~95% in the ME, AE, and CT cases, respectively. These numbers held true for all criteria pollutants and VOCs (including air toxics).
The BAU case represented increased emissions with no controls. Consequently, for this case, the CMAQ-simulated O3 concentrations along highways decreased because of NOx-titration. However, they resulted in the significant O3 formation in the NOx-limited outflow bordering the highway loop in Houston. The emissions control cases all resulted in O3 increases along the highways because of decreasing NOx saturation. However, they resulted in an O3 reduction in both regions enclosed by the highways and those in outflow areas. Simulated PM2.5 concentrations showed elemental and organic carbon hotspots along the highways, the magnitudes of which decreased with increasing control and fleet turnover. One important point was the removal of sulfate hotspots in 2040 resulting from the retirement of coal-fired EGU use. For the assessment of the health impact resulting from the change in O3 and PM2.5 concentrations, we found that the BAU case would lead to 122 additional premature deaths, while the ME, AE, and CT cases would prevent 114, 188, and 246 premature deaths, respectively. In addition, the prevented morbidities and economic cost (benefit) values generally mirrored premature mortality. Our analyses of the Houston Metropolitan area can shed light on the effects of mobile emissions control strategies in other coastal urban environments. Large urban cities can benefit significantly from reductions in PM2.5 pollution if local emissions from the transportation sector are controlled, while efficient O3 pollution reductions primarily occur in downwind areas.
The BAU case, with no change in emission factors, is technically not real since newly manufactured vehicles will follow the current latest emission standards, which are usually more stringent than previous standards. In addition, several state policies have implemented retrofitting (e.g. CARB, 2014; TCEQ, 2017) currently in place. These initiatives will result in reduction in future emission factors. In the other sense, the BAU case can represent revoking of advanced emissions standards and control policies, and hence, the "worst case" scenario. In terms of O3 pollution, the BAU case would not significantly change the aggregated numbers of premature mortality, but still have substantial impact on minor effects such as asthma exacerbation and school loss days. The CT case, assuming a fully turnover of the current fleet, will not be easy and require a range of strong policies. For example, current age distributions of vehicles vary by source type but almost all source types occupy notable fractions between 20 and 30 years (U.S. EPA, 2015c). Some heavy-duty trucks last more than 30 years and may turn from normal to high emitters (Liu et al., 2015). Hence, the BAU and CT cases are two extreme conditions. The actual future would be close to what exhibited in the ME case or close to those in the AE case with a stronger policy promoting electrification. Borkar et al. (2016) indicated significant retirement of fossil capacity and marked replacement of fossil fuel sources by renewables for the Houston eight-county area. The retirement of coal-fired power plants could provide an impetus to clean electrification in Texas; these efforts, however, might not be replicable everywhere. For example, according to the Emissions and Generation Resource Integrated Database (eGRID) (U.S. EPA, 2017c), about 67% of energy generation in the state of Ohio in the year 2014 was by coal, and the added load due to electrification could exacerbate an existing nonattainment problem. Hence, to gain a more comprehensive understanding of the overall effects of fleet electrification and the long-range transport of emissions outside of Houston and Texas, we would have to investigate several scenarios throughout the continental United States. It is also interesting to see how aggressive electrification would impact the air quality and health in the national level.
There are several limitations and uncertainties associated with this study. One important limitation is that a detailed microscopic traffic assessment and geodemographic (population density, transportation mode, land use) analysis is not included/fully considered in this study. For instance, the projected future mobile activity data provided by TTI are county-and vehicle type-specific, while the advent of autonomous vehicles has the potential to dramatically change the current transportation systems. Shared autonomous vehicles are expected to reduce travel distance and vehicle ownership, impacting on land use (e.g., parking infrastructure) and travel patterns (Shaheen and Cohen, 2013;Fagnant and Kockelman, 2014). Hence, the diurnal and spatial patterns of mobile emissions may also change. In addition, the population growth rates in the health impact model are at county-level, while the distribution of population could possibly change within each county. In this study, we made projections for on-road mobile and EGUs sectors, the emissions for other sectors were fixed to present levels. The meteorological conditions and chemical boundaries were also kept unchanged. These would affect the future total air quality levels (Bell et al., 2007a;2007b). Another uncertainty is that we applied fixed fractional rates of fleet turnover and control factor. The actual vehicle fleet composition may vary in the future and largely depend on policy, pricing, and technology. But the future emissions should fall between our four scenarios here. Also if technologies are not ready in 2040 to reduce non-exhaust emissions (e.g., PM2.5 emissions from brake and tire wear) from electric fleet, the corresponding change in PM2.5 concentrations and health outcomes would likely be smaller than the numbers reported in this study.
In the health impact analysis, the C-R relationships are derived from long-term air quality measurements and health records data, while only one month simulations were used here because the 1×1 km simulations were too computational expensive. Hence the health impact results may be over-or under-estimated if seasonal variation of air pollutants concentrations are considered. In addition, the uncertainties in emissions may impact the air quality and health outcomes. For instance, McDonald et al. (2018) reported the over-prediction of NOx mobile emissions in the National Emission Inventory. In our study, if the NOx mobile emissions are over-predicted in the base case, then the change in NOx emissions or concentrations between the base and future year cases would be smaller than the numbers reported in this study, resulting potential less change in O3 concentrations and subsequent overprediction of the health outcomes. While also as suggested by McDonald et al. (2018), reduction in mobile source NOx emissions could help reducing the number of high O3 days in the Eastern U.S. and meeting more stringent O3 standards in the future. In our cases, the fleet electrification could possibly lead to less emissions accumulation during the meteorological conditions favoring O3 production (e.g., land-bay/sea breeze recirculation, early morning low wind speeds), and potential less events of O3 exceedance. Overall, this study provides proof-of-concept of how the combined effects of a greening grid, emissions control, and fleet electrification can improve air quality and human health.

S1. A simplified flow chart of the air quality modeling system
Fig. S1. A simplified flow chart of the air quality modeling system.

S2.1. Fleet turnover with new and electric vehicles
This calculation was performed for the Aggressive Electrification scenario, which represents the highest electricity consumption for electric and hybrid vehicles and hence provides bounding estimates. The VMT data were fractionated into New, Electric, and Current types.
For light-duty "New" vehicles, we assume new technologies will all be hybrid, since major automobile manufacturers such as Volvo and Jaguar-Land Rover have already announced plans to build only hybrid vehicles post-2020. The United States Department of Energy (U.S. DOE)'s Energy Information Administration (EIA)'s Annual Energy Outlook of 2018 (AEO2018) provides the categories of hybrid (HEV), plug-in hybrid (PHEV), and all-electric vehicles (BEV) differing in range. According to the report, the proportion of electric vehicles with all-electric range (AER) 100, 200 and 300 in 2040 was projected to be 1:5:5 roughly, translating into overall fractions of 0.0636, 0.318 and 0.318 respectively (adding up to 0.7 that represents total "Electric" fraction). Similarly, the ratio of standard to plug-in hybrids was projected by the AEO to be roughly 3:1, resulting in 0.116 of hybrid and 0.034 of plug-in hybrid (adding up to 0.15 that represents total "New" fraction). The Argonne National Laboratory's Autonomie data further indicate PHEVs will be of four types, namely PHEV10, PHEV20, PHEV30 and PHEV40. In the absence of information, each of these assumed to have an equal share.
For heavy-duty diesel vehicles, in absence of adequate information on hybrid projections, we assume all new vehicles to be electric vehicles, resulting in 85% of electric vehicles in 2040.

S2.2. Energy consumption for hybrid and electric vehicles
Projections for electricity consumption (Wh mile -1 ) were taken from the Argonne National Laboratory's Autonomie model output (https://www.autonomie.net/). Their modeling involves projections all the way to 2040. The assumption is that a typical vehicle technology takes 5 years to reach the market once conceived in the laboratory. Hence, a laboratory year of 2035 is needed if we are considering a 2040 model year. A sample projection is illustrated in Fig. S2 for electricity consumption, indicating a linear trend from 2030 to 2045. Hence, the number for 2035 can be approximated linearly. Autonomie data covers the following vehicle classes: compact, midsize, small SUV, midsize SUV, pickup truck. These were mapped to MOVES categories as in Table S1 for electricity consumption per mile (Table S2). Autonomie considered three scenarios based on technology, this study considers the Medium case.
Since data for trucks were not available in Autonomie, these were taken from the Argonne National Laboratory's Alternative Fuel Life-Cycle Environmental and Economic Transportation (AFLEET) model (Burnham, 2017). The AFLEET data were reported originally in diesel-gallon equivalents, which were converted to Wh mi -1 using the assumption that one diesel gallon contains 145 MJ, and 0.2778 kWh are contained in 1 MJ, as illustrated by the U.S. DOE EIA's online calculator: (https://www.eia.gov/energyexplained/index.php?page=about_energy_conversion_calculator#dieselcalc ). The numbers are listed in Table S3.

S2.3. County-specific electricity demand
The calculation is illustrated with a simple example. For a given vehicle type (e.g. gasoline passenger car), the projected total VMT over the 8-county area is 1.5×10 8 miles in 2040. This is fractionated according to the distributions in Section S1.1, and multiplied by their respective electricity consumption values in Table S2, to obtain total electricity consumption for gasoline passenger cars. The process was then repeated for other vehicles to obtain a total.
We then summed over gasoline and diesel to get a total electricity consumption, which is 6.3×10 4 MWh. On comparing these numbers in context, the regional electricity generation over the 8county area is 7.5×10 7 MWh. Therefore, the electricity requirements for motor vehicles appear to be negligible compared to regional electricity generation. This could be due to that the Autonomie numbers represent improved technology and efficiency in 2040, which result in lowered consumption.   Fig. S4. The left is the description of the VOC species in the CB05 chemical mechanism (Yarwood et al. 2005). The right is averaged VOC speciation (in mole fraction) from on-road mobile emissions for the 8-county area, BASE Case.