Trends in life cycle greenhouse gas emissions of future light duty electric vehicles

Abstract The majority of previous studies examining life cycle greenhouse gas (LCGHG) emissions of battery electric vehicles (BEVs) have focused on efficiency-oriented vehicle designs with limited battery capacities. However, two dominant trends in the US BEV market make these studies increasingly obsolete: sales show significant increases in battery capacity and attendant range and are increasingly dominated by large luxury or high-performance vehicles. In addition, an era of new use and ownership models may mean significant changes to vehicle utilization, and the carbon intensity of electricity is expected to decrease. Thus, the question is whether these trends significantly alter our expectations of future BEV LCGHG emissions. To answer this question, three archetypal vehicle designs for the year 2025 along with scenarios for increased range and different use models are simulated in an LCGHG model: an efficiency-oriented compact vehicle; a high performance luxury sedan; and a luxury sport utility vehicle. While production emissions are less than 10% of LCGHG emissions for today’s gasoline vehicles, they account for about 40% for a BEV, and as much as two-thirds of a future BEV operated on a primarily renewable grid. Larger battery systems and low utilization do not outweigh expected reductions in emissions from electricity used for vehicle charging. These trends could be exacerbated by increasing BEV market shares for larger vehicles. However, larger battery systems could reduce per-mile emissions of BEVs in high mileage applications, like on-demand ride sharing or shared vehicle fleets, meaning that trends in use patterns may countervail those in BEV design.


Introduction
Transportation comprises 28% of US greenhouse gas (GHG) emissions, 60% of which come from light-duty vehicles (LDVs) (US Environmental Protection Agency, 2018). While a multipronged approach is needed to achieve deep reductions in transportation GHG emissions, rapid and extensive deployment of battery electric vehicles (BEVs) is viewed as a crucial part of nearly all strategies (Alexander, 2015a;Meszler et al., 2015;Sperling, 2018). BEVs are typically referred to as zero emissions vehicles (ZEVs) because they eliminate tailpipe pollution.
However, as with other de-carbonization policies for the transport sector, such as those that promote biofuels, a life cycle perspective is required to understand the actual mitigation achieved by ZEVs, since emissions are not eliminated, but rather shifted upstream in the fuel cycle (to the power plant) and potentially increased in the vehicle production supply chain. BEVs can also have considerable variability in life cycle operation emissions given the heterogeneity of electricity grids over space and time (Cerdas et al., 2018;Yuksel and Michalek, 2015).

Numerous life cycle-based studies have been conducted with the goal of verifying if BEVs
(including plug-in hybrid electric vehicles (PHEVs)) achieve real reductions in emissions relative to internal combustion engine vehicles (ICEVs, including hybrid electric vehicles (HEVs)). These studies suggest that GHG emissions associated with energy for BEV operation (i.e. production of electricity) can be 44% -80% of BEV LCGHG emissions. For non-operation GHG emissions, energy required for manufacturing of LIBs is the primary driver of increased GHG emissions relative to ICEVs (Peters et al., 2017). Uncertainty about battery manufacturing and a lack of primary data have contributed to a wide range of results for GHG emissions associated with battery production (Ambrose and Kendall, 2016;. Moreover, given the growth in BEV sales, the evolution of BEV designs and model availability, and declining prices for traction batteries (Nykvist and Nilsson, 2015), previous life cycle assessments (LCAs) may not be representative of current and future BEV performance, vehicle specifications, or patterns of use.

Review of Literature and Relevant Data
A review of previous LCAs (here we use the term LCA to refer both to comprehensive LCAs that track a suite of environmental impacts as well as those that narrowly assess GHG emissions), a selection of which are summarized in Table 1, shows that most studies used the early generations of the Nissan Leaf as the exemplar BEV (Archsmith et al., 2015;Graff Zivin et al., 2014;Hawkins et al., 2013;Majeau-Bettez et al., 2011;. Because of this, most previous LCAs have used similar assumptions, including the ~24 kWh battery capacity and efficiency-oriented compact vehicle design. Many of the earliest LCA studies of BEVs found that emissions from the electricity grid used for charging were the most significant contributor to life cycle CO2e emissions from BEVs (Hawkins et al., 2012;Michalek et al. 2011). Justifiably, more recent studies have focused on interactions of BEVs and the electricity system, examining the consequential effects of replacing ICEVs with BEVs, and the intersection of charging strategies with the marginal dispatch decisions of electric utilities (Archsmith et al., 2015;Jenn et al., 2016;Yuksel and Michalek, 2015). At least one study has considered the effect of battery range and vehicle size on BEV performance (Ellingsen et al., 2016). They found commensurate increases in LCGHG with increasing battery and vehicle size and, similar to previous studies, found that electricity grid carbon intensity determined the preference of BEV vehicles over their conventional fossil fuel counterparts.

Studies of BEVs and Gasoline Vehicles
While previous studies provided valuable insights about the life cycle performance of vehicles and the importance of electricity grid emissions (whether modeled as marginal or average emissions), the majority of these studies reflect outmoded assumptions about BEV vehicle designs and did not reflect trends in the BEV market. A review of US BEV sales between 2012 and 2018 shows a marked shift towards significantly higher capacity batteries, longer vehicle ranges, and an increasing preference for high performance and luxury BEVs. The combined effect of these two trends is evident in Figure 1, which shows the US sales-weighted average annual increase in BEV battery capacity of 6.5 kWh per year between the first quarter of 2012 and the second quarter of 2018, reaching 74 kWh by the second quarter of 2018. As the market for BEVs has grown, so too have the number of BEV models available. Instead of the efficiencyoriented compact passenger vehicle, the fastest selling BEV in the US has become the leader in the luxury sedan segment (Alternative Fuel Data Center, 2018). Sport-utility BEVs have emerged as an important market segment with several major vehicle manufacturers launching cross-over style BEVs (Gale, 2018).
Two important trends in personal mobility are also changing the use-cases for BEVs: one, the increased use of and participation in on-demand ride sharing services; and two, increased reliance on automated and connected vehicle technologies to replace human driving activities (Greenblatt and Shaheen, 2015). While the net effects of these trends on vehicle travel is still unknown, the emergence of ride-hailing services like Uber and Lyft are having significant impacts on traditional modes (e.g. transit) and historical patterns of mobility (Clewlow and Mishra, 2017;Hall et al., 2018). Based on early research, individual shared or automated vehicles could generate three to four times the comparable annual VMT of a conventional (private) passenger vehicle (Fagnant and Kockelman, 2014;Gurumurthy and Kockelman, 2018;Loeb et al., 2018). Vehicles participating in ride-hailing services can also experience significant mileage from return links, also known as dead-heading (Henao, 2017 Keoleian (2015)). Many researchers have also highlighted the problem of data availability in the context of prospective assessments or emerging technology assessments, noting not only the challenge of modeling the performance of a technology not yet in the market, but also the lack of temporally appropriate background data for prospective assessment (e.g. Hetherington et al. 2014;Arvisson et al. 2018). To create a practicable scope of assessment, this study focuses only on trends in electric vehicle design with respect to performance characteristics and battery capacity, and considers changes to only a few background systems (e.g. the electricity grid).

FIGURE 1 BEV sales and battery capacities in the US
The combined effects of larger battery capacity; a shift towards large, high-performance BEV models; and the increased use of BEVs in high-mileage applications may challenge some of the widely accepted conclusions of earlier BEV LCAs, namely the small contribution of vehicle production-related emissions to life cycle emissions and that in many parts of the US (and in regions throughout the world) BEVs provide GHG mitigation benefits (albeit sometimes small) relative to ICEVs. This observation led to the following research questions explored in this study: ( (2) What is the combined effect of vehicle design trends and technology and electricity grid evolution on the LCGHG emissions intensity of BEVs?
(3) How will these trends effect future emissions rates of BEVs, particularly in high-mileage applications like shared ride fleets

Goal and Scope
This study aimed to quantify the LCGHG emissions of three archetypal future BEVs that reflect the changing BEV market, as described below: Archetype 1 -An efficiency-oriented compact vehicle (EOV), based on the Chevrolet Bolt.
Archetype 2 -A high performance luxury sedan (PLS), based on the Tesla Model S P100D.
Archetype 3 -A high performance SUV (PSUV), based on the Tesla Model X P100D.
For each vehicle archetype, the study considers how future changes in vehicle design, battery performance, changing electricity grid, and annual mileage will affect the total LCGHG For each vehicle scenario, we evaluate a set of 2025 models with improved battery systems (Table 2). We then compare this to both current market BEVs, as well as a set of 2025 models with increased battery capacity and travel range (Long Range or LR). Vehicle scenarios are evaluated across a set of use-phase scenarios reflecting differences in travel behavior, vehicle life, and electricity generation. The model includes both the operation and non-operation stages of the vehicle life cycle.
The vehicle life cycle is divided in two phases; the vehicle phase, which includes vehicle production and disposal, and the operation phase. The vehicle phase is broken down into the battery system and the rest of the vehicle, referred to as the glider. The end-of-life (EOL) stage includes disposal and recycling of the glider. Disposal and/or recycling of the traction battery is not included because of uncertainty in how batteries will be managed in the future, particularly as many more batteries are retired and either recycling networks or second life uses emerge.
Use-phase emissions for BEVs are then estimated as a function of vehicle energy efficiency and the emissions associated with electricity production and delivery. Two sets of travel scenarios were applied to the different vehicle models shown in Table 2: 1. A privately-owned vehicle in an average US Household (referred to as the AVE scenario) 2. A service vehicle deployed in an urban, ride-hailing fleet (referred to as the SAV scenario) To capture regional variability, changing fuel sources, generation technologies, and policy in the electricity system, a range of electricity generation forecasts were modelled for both California and the US region from the period 2017 to 2025. The electricity generation scenarios are discussed further in a later section.

LCI Inventory Model
The life cycle inventory (LCI) model tracks only energy consumption and GHG emissions. A three part LCI model was developed to estimate the required inputs of energy and raw materials and resulting emissions: part one evaluated the production of the vehicle glider body and balance of systems (the glider model); part two evaluated the production of the battery system; and part three evaluated the generation of electricity supplied to charge the vehicle. 1

Glider Model
The glider model examined the life cycle emissions of the vehicle without the battery, which included raw material acquisition and refining, processing, assembly and disposal. The reference LCI data for this model was acquired from the Greenhouse Gases, Regulated Emissions, and Other assumptions included the mass and number of replacements for fluids and tires, also acquired from the GREET 2 model. Further, because electricity use does not constitute a large portion of total energy use and resulting emissions in this phase, time dependence of the electric grid was not considered in the glider model-meaning that a vehicle produced in the future is modeled using the same electricity grid LCI as those produced today. For both 2018 and 2025 scenarios, glider material composition as well as per-mass emissions are assumed to be the same.
And since no light-weighing was assumed, glider masses also remain the same. The baseline ICEV car, SUV, and HEV scenarios presented for comparison are taken from the default vehicle set in GREET 2. The resulting estimates for the material balance of the vehicles, the average energy input for assembly processes, and further details on the vehicle model can be found in the Supporting Information S3.

Battery Production
Battery production LCIs were developed using the model described in Ambrose and Kendall  Kendall (2016). All vehicle scenarios are assumed to use a lithium nickel manganese cobalt (NMC) battery chemistry. Variations of NMC have emerged as the dominate cathode chemistry for most light duty applications owing to its high specific power (Olivetti et al., 2017). The composition of lithium ion battery (LIB) packs can vary due to the type of cells used, thermal management systems, and structural elements. There is also considerable uncertainty in estimating the energy required for assembling LIB cells owing to limited, poor quality data (Peters et al., 2017). We considered several futures for battery design, production processes, and key inputs through a scenario based sensitivity analysis. These results, the normalized average material composition for each battery pack, assembly emissions estimates, as well as more discussion on the battery production model is included in the Supporting Information S4.
The baseline assumption is that no battery replacements are required over the course of a vehicle's lifetime. This assumption and the conditions where battery replacement is likely to be needed is discussed in Section 3.1.    Table 3 shows the assumed curb weight and key vehicle specification inputs by vehicle scenario.

Vehicle Energy Demands
The aerodynamic and motor specifications are held constant across each class of vehicle modeled. As explained in the section on Glider model, curb weights vary according to battery system improvements and battery sizing.
It is widely expected that recent developments in LIB technology will enable battery packs with nearly double the energy density of early EV batteries. Use of higher capacity cathodes, more efficient thermal management, improved electrolytes and anode materials could increase battery specific energy from today's ~130 Wh/kg to over 250 Wh/kg by 2022 (Elgowainy et al., 2016).
The current (2018)  increase accordingly. Additional review of the BEV assumptions and discussion is provided in the Supporting Information S5.
The ICEVs included in the study for comparison are drawn directly from GREET 2, and include both emissions from fuel production (i.e. WTP) and fuel combustion (i.e. PTW). The ICEV WTP and PTW emissions rates were estimated from the default ICE sedan, SUV, and HEV scenarios and VMT assumptions in GREET 2. The average fuel economies for these scenarios are 34 MPG for the sedan, 24 MPG for the SUV, and 42 MPG for the HEV respectively, while upstream emissions from fuel production as a share of combustion emissions (i.e. WTP/PTW) is 0.24 to 0.27.

Vehicle Miles Travelled (VMT)
Automotive LCAs commonly rely on an assumption of fixed or average lifetime mileage, often based on data from industry associations or anecdotal data (Weymar, 2016). In reality, the total miles travelled by the vehicle lifetime (LifetimeVMT) is driven by two phenomena (Eq. 2): one, the scrappage rate (M), which is the probability or fraction of a given model year's vehicles

Charging
BEVs are likely to utilize a range of private or public charging infrastructures with different power levels for charging events, which could impact the efficiency of refueling the vehicle (Smart and Schey, 2012;Tal et al., 2014). Sears et al. (2014) collected data on charger efficiency for a range of charging power levels and climate conditions from a small sample of Nissan Leaf and Chevy Bolt drivers; the authors found efficiency ranged 83.8% to 89.4% for Level 1 vs 2 charging events. There are much more limited data is available for the efficiency of high power chargers. It is likely that any variability in BEV emissions rate attributable to variation across charging infrastructures is less than that due to climate, driving distance, and other factors (Taggart, 2017). In this study, an average efficiency of = 86% is used for all scenarios, and the sensitivity of results to this assumption is explored in the discussion.

Electricity Generation
LCAs  only 8% to 12% of LCGHG emissions for ICEVs are attributable to vehicle production, production emissions were estimated to contribute 30% -66% of per mile emissions for BEVs.
Production of the battery system contributed 28% -51% of vehicle production emissions for BEVs, and 11% to 35% of overall per mile emissions. to 367 gCO2e/mile), due to higher total lifetime mileage, which led to a lower contribution from vehicle and battery production on a per-mile basis as well as more vehicle miles accumulated with lower LCGHG intensity electricity. Mileage Scenario 3, in which vehicle lifetime is fixed at 12 years for all vehicles, resulted in less than a ±1% change in all non-SAV applications, as shown in Table S8.1 of the Supporting Information.  In these high mileage applications, it is also expected that key vehicle systems will require additional replacement due to excessive wear. The results reported for the SAV scenarios assume replacement of vehicle battery based on expected lifetimes. Battery systems are assumed to be replaced after delivering a fixed number of equivalent charge and discharge cycles, and the estimates in Figure 3 for BEV SAVs assume an average 1 to 1.5 battery replacements over the average 12 year vehicle life. Vehicle powertrain, chassis, and other systems were not assumed to experience additional replacements as a function of mileage. An expanded results section, including a full accounting of results from the carbon tax and SAV scenarios is include in the supporting information (S8). The service life of the battery is discussed further in the next section.

Battery Replacement and Vehicle Lifetimes
Battery cycle life is generally defined by the total number of times a battery can deliver its energy storage potential in a particular discharge program (Barré et al., 2014;Fortenbacher et al., 2014;Han et al., 2014), thus the service life will vary under different duty cycles and operating conditions. The effective cycle life is highly dependent on the utilization of storage potential and the rate of discharge. A common metric or measurement of battery performance is cycles to 80% depth of discharge (DOD), or 80% of the battery energy storage potential. Cycles to 80% DOD is also convenient as utilization of the battery near the maximum and minimum of the battery potential are associated with accelerated battery degradation. Many battery systems are managed to prevent discharge below or charging above a certain threshold to prevent damage to the battery system. While early lithium ion cells might only deliver several hundred cycles before

Electricity
Emissions generated during the vehicle use-phase from producing electricity to charge the vehicle are on average more than 50% of LCGHG emissions. A key uncertainty in estimating use-phase emissions for BEVs stems from variability in the emissions rate for delivered electricity. The effects of BEV efficiency on per mile emissions have also been poorly addressed in many previous studies due to the limited types of vehicles evaluated. Figure

Discussion
Increasing BEV battery capacities could have mixed impacts on the life cycle emissions rate of grid-tied BEVs and the GHG abatement from a transition away from gasoline-powered vehicles. examined trends in BEVs, as we understand them today, and extend them into the future. In addition, ICEV technologies were treated as static, meaning that improvements in ICEV technologies and fuels were not considered.

Section S3. Vehicle Production
To model the emissions of the glider production, the material composition of the glider and its mass along with life cycle inventory of those materials is required:.
The life cycle inventory used to model LCGHG emissions associated with vehicle production, assembly, and disposal was acquired from the GREET 2 model (Argonne National Laboratory, 2017a). Their data on material acquisition and transformation, vehicle assembly, and vehicle disposal was combined with glider mass and composition to estimate emissions. In this study, the battery system was modelled separately from the rest of the vehicle, referred to as the glider.
Hence the glider mass is calculated by subtracting battery mass from the curb weight. The material compositions for the Leaf (2012), PLS and the PSUV vehicle scenarios are based on material composition used in a similar study (UCS 2015) which builds off of material data used in GREET 2 model. The material composition for the EOV scenario is from the vehicle teardown performed on Chevrolet Bolt by Munro associates.
The material composition of the glider and mass used for each of the four modeled vehicle scenarios can be seen in Table S3.1. Additional considerations included material transformation, fluid use, assembly, and disposal, which were also acquired from GREET 2 model. Fluids are included in the body and powertrain material life cycle stage of this study's model, and in EVs include brake fluids, powertrain coolant, and windshield fluid; sedans and SUVs were assigned different sets of lifetime fluid use, with the latter having higher fluids use. All vehicles were given identical assembly and disposal impacts, where the energy use was 11.57 mmBTU and 3.26 mmBTU respectively. Note that the modeled emissions may underestimate true impacts, as the life cycle emissions of the approximately 3% of "other" materials was not accounted for. Additionally, since electricity does not play a major role in this phase, time dependence of the electric grid was not included.

Section S4. Battery Production
The battery production model examined the cradle to gate of the battery life cycle, which included emissions from raw material extraction and refining, production, and assembly. An existing tool, the Battery Performance and Cost Model (BatPaC) constructed by Argonne National Laboratory, was used as the basis for the battery production model. BatPaC is based on a robust study of the material properties of LIB electrode and packaging materials, as well as battery pack design and production. BatPaC estimates the cost and composition of the LIB pack systems; in prior work (Ambrose, 2016), we connected these these outputs to material life cycle inventory data to estimate the GHG intensity of battery production processes. BatPaC offers the capabilities to compare the performance of different LIB cathode materials, however nickel rich cathode compounds NMC (e.g. 622 and 811), are being predicted to dominate light duty automotive applications (Curry, 2017). Table S4.1 summarizes the key parametric assumptions relating to the battery pack design, i.e. pack size, mass, power output, and cell and module capacity. The scenarios were developed based publicly available data on current models of archetypal vehicles described in main text. The resulting breakdown of key materials are summarized in Table S4.2. Material LCIs were then obtained from the GREET 2018 model, and used to estimate the total energy and global warming potential for battery material production (measured in CO2 equivalents, or GHGs). We also conducted a sensitivity analysis on assumptions about battery assembly energy requirements (as measured by the kWh of energy input per kWh of usable storage) and pack energy density for the future vehicle case (Table S4.3). .  Table S4.3 shows the results of the scenario based sensitivity analysis of battery production energy and GHG emissions. Under the high assembly energy scenario, total energy requirements and GHG emissions more than doubled. While the efficiency of production processes increases significantly, those gains are not sufficient to offset the increases in battery capacity. Energy inputs and GHG emissions from battery assembly are primarily attributable to environmental controls and formation cycling. We assumed a constant inventory for battery assembly energy based on electricity generation for industrial purposes in South Korea. If the primary energy source for battery assembly was changed, this could significantly impact the emissions attributable to battery assembly energy inputs.

Section S5. Vehicle Energy Demands
FASTSim is a system analysis tool by NREL to compare the drivetrain performance. The model was first verified by modifying the inputs for three vehicles of our focus and cross checking the resulting fuel economy values with the 2018 values reported by the EPA. The vehicle parameter inputs are provided in Table S5.1

Section S6. Annual Vehicle Miles Travelled (VMT)
Two sets of scenarios for vehicle travel were developed: one, representing primary use in a personal passenger vehicle application (US Department of Transportation).; and two, representing use in a shared on-demand or potentially automated ride-hailing fleet. In all scenarios, annual VMT decreases as the vehicles age due to a variety of factors. Table S6.1 shows the estimated annual mileage function for each scenario obtained from the regression of annual VMT on vehicle age within the NHTS data. The lifetime vehicle miles travelled presented in Table S6.1 were estimated using the survivability data for cars, SUVs, and taxis obtained from Jacobsen et al. (2015), and Bishop et al. (2016) respectively. Table S6.2 summarizes the survivability data used. Finally, the total lifetime VMT for each scenario is provided for both passenger and SAV scenarios in Table S6.3

Section S7. Electricity Generation
This section provides the complete results of the electricity generation analysis and the resulting forecast for grid carbon intensity. The study considered two regional scenarios: the California subset of the WECC region (CAMX) and a US national average. The study also considered two policy scenarios: a business as usual case and a carbon tax scenario with a $25 dollar per ton cost of carbon. The Annual Energy Outlook 2018 defines the Reference case in which: population (including armed forces overseas) grows by an average rate of 0.6%/year, nonfarm employment by 0.7%/year, and productivity by 1.6%/year from 2017 to 2050. The real gross domestic product increases by 2.0%/year from 2017 through 2050, and growth in real disposable income per capita averages 2.2%/year (U.S. Energy Information Administration, 2018).
For all scenarios, the study considered a time horizon from 2018 to 2050. Data on the net electricity generation by year by fuel source was obtained from the Annual Energy Outlook created by the Energy Information Administration. The AEO forecast is based on outputs of the National Energy Model, a large scale economic equilibrium model of energy supply and disposition (Gabriel et al., 2001).
The average net generation by fuel source is provided for a subset of years in Table S7.1. The average generation by fuel source data was combined with the life cycle emissions inventory data to estimate the emissions rates by year. For each fuel source, a regionally representative LCI was estimated using data from the GREET 1 model (Argonne National Laboratory, 2017b). Table S7.2 shows the estimated LCIs by fuel source and scenario. The final row of the table shows the estimated total greenhouse gas emissions of each kilowatt hour provided in carbon dioxide equivalents. A 100 year global warming potential is assumed, with characterization factors taken from the IPCC AR5.  Finally, the estimated average carbon intensity of electricity generation for each year is provided in Table S7.3.

Section S8: Full Results
This section provides a table view of the complete results of the final GHG estimates.
Beginning on the next page, table S8.1 contains per mile GHG emissions attributable to vehicle and battery for each vehicle design scenario. The results in table 8.1 for per mile emissions attributable to production of vehicle and battery systems use the survivability method for estimating lifetime vehicle miles. The results are moderately reduced in the VMT scenario due to the assumed higher lifetime vehicle miles travelled. The phase column corresponds with the key categories of emissions in producing the vehicle and battery system. The survival and VMT methods result in the same estimated emissions rate for conventional ICE vehicles as the ICE vehicle is the basis for the VMT method used in the BEV cases.  Table S8.3 provides the total results, which are the sum of the vehicle and battery emissions with the use phase emissions. As such, the total results are presented by grid scenario and service life in years. Table S8.3 makes clear the key trend, namely the increasing share of production emissions in life cycle emissions and per mile emissions for passenger vehicles.