Our model consists of two main components – a fleet model that determines the number and types of vehicles in each state, and an emissions model that determines the life cycle greenhouse gas emissions of the vehicle fleet.
Fleet model
The fleet model is based on the historical vehicle stock45, projected vehicle sales46, and vehicle scrappage rates26 (Fig. S1-S3). The number of vehicles in the fleet (and of each model year) is determined by the number of vehicles in the fleet in the year prior, plus projected sales, minus the scrapped vehicles determined by the vehicle survival curves.
$${N}_{y,p,c}={N}_{y-1,p,c}+Sale{s}_{y,p,c} -Scrappe{d}_{y,p,c}$$
where y is the vehicle year, p is the vehicle powertrain, and c is the vehicle class. For powertrains we use ICEVs and EVs, where ICEV represents the sum of ICEV-gas, ICEV-diesel, CNG, FCV, HEV, and PHEVs, and EV represents the sum of 100-mile, 200-mile, and 300-mile range EVs. For classes we use cars and trucks, as defined in the 2022 Annual Energy Outlook sales projections46.
State fleets
To set the initial conditions for the state fleets, we need the percentage of car sales that were electric in 2021 and the percentage of truck sales that were electric in 2021 in each state. We approximate this data by first taking the percentage of LDV sales that were electric in each state, \({\frac{EV Vehicles}{Vehicles}}_{s}\), from the Alliance for Automotive Innovation47, the number of cars and trucks relative to the total number of vehicles in each state, \({\frac{Cars}{Vehicles}}_{s}\) and \({\frac{Trucks}{Vehicles}}_{s}\), from vehicle registration data, and the number of EV cars and trucks compared to the total number of EVs nationally, \(\frac{EV Cars}{EV Vehicles}\) and \(\frac{EV Trucks}{EV Vehicles}\), from 2021 sales data, and solving for our desired initial values, \({\frac{EV Cars}{Cars}}_{s}\) and \({\frac{EV Trucks}{Trucks}}_{s}\), in each state using the system of equations shown below:
$${\frac{Cars}{Vehicles}}_{s}*{\frac{EV Cars}{Cars}}_{s}+ {\frac{Trucks}{Vehicles}}_{s}*{\frac{EV Trucks}{Trucks}}_{s}={\frac{EV Vehicles}{Vehicles}}_{s}$$
$$\frac{{\frac{EV Cars}{Cars}}_{s}}{{\frac{EV Trucks}{Trucks}}_{s}}=\frac{\frac{EV Cars}{EV Vehicles}}{\frac{EV Trucks}{EV Vehicles}}$$
While imperfect, these data are combined to give estimates of the percentage of cars that are electric and the percentage of trucks that are electric in 2021, in each state.
We assume that the adoption rate is the same for cars and trucks and is the same in every state. The initial condition (percent electric in 2021) leads to different outcomes for cars and trucks and within each state as seen in Fig. 1a,b. This also accounts for the fact that car-to-truck ratios vary greatly from over 2 in Washington DC to less than 0.5 in Alaska. We solve for the adoption rate, \(r\), using a simple logistic function such that across all 50 states the total percentage of EV sales is exactly 50% in 2030:
$${P}_{c, s}\left(2030\right)=\frac{K*{P}_{0,c,s}{e}^{{r}\left(2030\right)}}{K+{P}_{0,c}\left({e}^{{r}\left(2030\right)}-1\right)}$$
where \({P}_{c, s}\left(2030\right)\) represents the percentage of cars that are electric in each state, s, in 2030. The carrying capacity, K, is 1 (corresponding to a maximum sales percentage of 100% electric vehicles). The initial condition, \({P}_{0,c,s}\), is the percentage of cars sales, c, that were electric in each state, s, in 2021.
$${P}_{0,c,s}={\left(\frac{EV Cars}{Cars}\right)}_{s}$$
This is also done with trucks in every state:
$${P}_{t,s}\left(2030\right)=\frac{K*{P}_{0,t,s}{e}^{{r}\left(2030\right)}}{K+{P}_{0,t,s}\left({e}^{{r}(2030}-1\right)}$$
where \({P}_{t, s}\left(2030\right)\) represents the percentage of trucks that are electric in each state, s, in 2030. The initial condition, \({P}_{0,t,s}\), is the percentage of truck sales, t, that were electric in each state, s, in 2021.
$${P}_{0,t}={\left(\frac{EV Trucks}{Trucks}\right)}_{s}$$
We then take the summation of the percentage of cars that are electric in each state multiplied by that state’s proportion of the total number of cars in the country, and the percentage of trucks that are electric in each state, multiplied by that state’s proportion of the total number of trucks in the country, and set that equal to are sales target of 50% (or any other goal):
$$\sum _{s=1}^{51}{P}_{c, s} \left(2030\right)*\left(\frac{Car{s}_{s}}{Cars Nationally} \right)+ {P}_{t,s}\left(2030\right)*\left(\frac{Truck{s}_{s}}{Trucks Nationally }\right)=50\%$$
This assumes that each state’s proportion of the total number of vehicles in the country is constant throughout the study period. This system of equations can then be solved for the percentage of car and truck sales that are electric in each state in 2030, \({P}_{c, s} \left(2030\right)\) and \({P}_{t,s}\left(2030\right)\), and the required growth rate to reach those values, \(r\).
Emissions model
Once the number of vehicles (of each class, powertrain, and age) in each state has been determined, the use phase emissions can be calculated by multiplying the number of vehicles by the VMT (per vehicle), and by the greenhouse gas intensity of travel (per mile). These factors (number of vehicles, VMT, fuel economy, and grid intensity) can vary based on powertrain, vehicle class, location, and year.
$$LDV Fleet Use Phase Emissions={N}_{p,c,l,y} *VM{T}_{p,c,l,y}*{I}_{p,c,l,y}$$
where \(N\) is the number of vehicles, \(VMT\) is the vehicle miles traveled, I is the GHG intensity of travel, \(p\) is the powertrain, \(c\) is the vehicle class, \(l\) is the location and \(y\) is the year. Not every component of the equation varies with all four indices. For example, we use VMT profiles that vary based on vehicle class and vehicle age but are the same across powertrains and locations (Fig. S4). For ICEVs the GHG intensity of travel is determined by the life cycle carbon intensity of the fuel, \(C{I}_{F}\), in kg CO2e/gallon and the fuel economy, \(F{E}_{ICEV}\), in miles/gallon.
$${I}_{ICEV}=\frac{C{I}_{F}}{F{E}_{ICEV}}$$
For EVs the GHG intensity of travel determined by the fuel economy, \(F{E}_{EV}\), in Wh/mile and the grid emissions factor, \(EF\), in kg CO22e/MWh, with appropriate unit conversions.
$${I}_{BEV}=F{E}_{EV}*EF$$
Fuel economy
Fuel economy data comes from the VISION Model48. We take the weighted average (by market share) of 100-mile, 200-mile, and 300-mile range EVs to determine the annual fuel economy of EV cars and EV trucks for each year. We divide the EV fuel economy, in Wh/mile, by 0.88 to account for charger efficiency. We use a similar weighted average for ICEVs, which consists of ICEV-gas, ICEV-diesel, CNG, FCV, HEV, and PHEV cars and trucks (Fig. S5). For both ICEV and EV fuel economies we use the on-road correction factors from the Vision Model48. We also take a weighted average carbon intensity of the fuels for each year, corresponding to the different fleet mixes of each year, using data from GREET49. This includes combustion emissions and upstream emissions.
Grid emissions
We use grid emissions projections from NREL’s Cambium model27 for each state, as seen in Fig. 1 (using the methods of Vega-Perkins et al., to obtain values for Alaska and Hawaii)50. We use two different emissions scenarios – a business as usual scenario (called the midcase within Cambium), and a 95% decarbonization by 2035 scenario (compared to 2005 levels). We use the combined combustion and upstream GHG emissions of electricity demand (not generation) in each state (CO2, CH4, and N2O, 100-year GWP). Transmission and distribution losses are included in the Cambium values.
Vehicle production emissions
We obtain vehicle production emissions for model year 2020 and model year 2030 vehicles from Woody et al., (2022)11. We translate the three vehicle classes in that study (midsize sedan, midsize SUV, and pickup truck), into the two categories used in here (car and light truck), based on the 2021 sales ratios of the five categories used by the EPA16 (Fig. S6, Supplemental Note 1). We use linear interpolation to calculate vehicle production emissions specific to each year. Note that vehicle end-of-life emissions although relatively small are also included in the overall vehicle production emissions.
Attribution analysis
For the base case, vehicle electrification reaches 50% of sales in 2030, grid carbon intensity declines from 450 to 340 kg CO2e/MWh between 2022 and 2030 (“business-as-usual”), ICEV average new vehicle fuel economy improves by approximately 3 MPG from 2022 to 2030, EV average new vehicle fuel economy is essentially unchanged from 2022 to 2030, the percentage of vehicles sold that are trucks increases from 67–72%, and the total number of vehicles sold annually declines by 1%. The VMT per vehicle annual schedule, which includes declining VMT as the vehicle ages, does not change throughout the study period. To construct Fig. 4, the emissions were calculated with a static truck sales percentage (67%), static EV sales percentage (approximately 4%), and static grid emissions (450 kg CO2e/MWh). This shows the emissions reduction that will occur without these trends (i.e., due to fleet turnover). Fleet turnover captures the impact of improved fuel economy, the small change in the overall vehicle fleet size and most significantly, the replacement of older vehicles. The impact of truck percentage and electrification and grid improvement are calculated sequentially. Electrification and grid improvement are combined, as these trends interact strongly, as discussed in the Results. Reaching 50% EV sales increases the emissions due to vehicle production. The increase in vehicle production emissions shown in Fig. 4 represents the additional emissions from reaching the 50% EV sales target, relative to the static scenario in which EV sales remain at 4%.