Pathways to zero emissions in California's heavy-duty transportation sector

• Our main data source is CARB's EMFAC online emissions and fleet database [27]. The emissions component of the tool includes historical data and future projections of vehicle stock, VMT, and both GHG and criteria air pollutant emissions. We extract statewide data from 2019 to 2045, broken down by vehicle type, chassis model year, and fuel type. The fleet database contains historical data on vehicle fleet composition. We extract data on the number of vehicles of each type, chassis model year, and fuel type by census block group for 2019. Hereafter, 'model year' refers to chassis model year.
• We take current CO₂ emissions factors for electricity used for transportation from CARB's low carbon fuel standard [28].
• We use demographic data from the 2019 5 yr Estimates from the American Community Survey [29] at the census tract level to conduct our demographic analysis.
• We use consumption-based PM_2.5 concentration increase factors from Hennessy et al [30] to perform our air quality analysis for electric vehicles.

We model the evolution of the vehicle fleet from 2019 to 2045 using a stock and flow model. The model relates vehicle stock in the next year to vehicle stock, retirements, and sales in the current year. Vehicle stock is broken down by age, vehicle type, and fuel type. The relationship is defined in equation (1), where t represents the current year; t + 1 represents the next year; Q_t+1,m,v,f represents the stock of vehicles of type v, fuel type f, and model year m in year t + 1; Q_t,m,v,f represents the stock of vehicles of type v, fuel type f, and model year m in year t; R_t,m,v,f is the retirement of vehicles of type v, fuel type f, and model year m in year t; and S_t,m,v,f is the sales of vehicles of type v, fuel type f, and model year m in year t,

$$\begin{equation}{Q_{t + 1,m,v,f}} = {\text{ }}{Q_{t,m,v,f}} - {\text{ }}{R_{t,m,v,f}} + {\text{ }}{S_{t,m,v,f}}.\end{equation} \tag{ 1 }$$

We fix the total stock for each vehicle type in each year across scenarios, to allow for comparisons to be made across scenarios. We use total stock by vehicle type from CARB's projected stock in the EMFAC emissions database [27] as the total stock. CARB's vehicle stock forecasts are based on national vehicle sales growth trends from the annual energy outlook paired with county-level estimates of VMT from the California statewide travel demand model (CSTDM) [31], which is a multi-modal activity-based travel demand model [32]. Baseline projections include an increase in long-haul freight transport represented by an increase in heavy-heavy duty vehicle population and VMT (see figures S18 and S19 for BAU stock and VMT projections). Anticipated modal shifts are implicitly included via the CSTDM's multi-modal design.

Retirements are determined using survival profiles extracted from EMFAC, which specify the percentage of vehicles of a given vehicle type that survive to a specified age (see figure S13 for survival profiles for each vehicle type). We use CARB's projected vehicle stock by vehicle type for vehicle years with a model year of 2020 to calculate the survival profiles. For simplicity, we assume that the survival profile for a given vehicle type does not change throughout time, and that the survival profile for a given vehicle type is the same for all fuel types. Future work could incorporate variations in survival profiles for different fuel types and model years.

Sales are then calculated as the sales needed to make up for the loss in stock due to retirements to achieve the projected stock in the next year. We do not explicitly model the sale of used vehicles, and assume that all sales are new vehicles (age = 0). For each vehicle type, we specify the breakdown of sales by fuel type in each year as a percentage of total sales.

We include ten vehicle types in our study: two classes of light HDVs ('LHD1' and LHD2"); local medium HDVs ('T6'); long-haul medium HDVs, operating both within California and in other states ('T6 OOS'); local and long-haul heavy HDVs ('T7' and 'T7 OOS'); heavy HDVs operating in ports ('T7 Port'); buses, motorcoaches ('MC'); and motorhomes ('MH'). See table S1 for details.

To model ZEV sales mandates, we adjust the breakdown of sales by fuel type in each year. We assume that the percentage of ZEV sales increases linearly from 2019 to the year the sales mandate is implemented. We assume the relative distribution of the other fuel types remains the same. For simplicity in our analysis, we assume that all ZEV sales are battery electric vehicles, but the model has the capability to specify a split between hydrogen and electric vehicles. Future work could include scenarios with different types of ZEVs.

To model accelerated retirement policies, we add a term, E_t,m,v,f representing early retirements of vehicles of type v, fuel type f, and model year m in year t to equation (1), resulting in equation (2),

$$\begin{equation}{Q_{t + 1,m,v,f}} = {\text{ }}{Q_{t,m,v,f}} - {\text{ }}{R_{t,m,v,f}} - {\text{ }}{E_{t,m,v,f}} + {\text{ }}{S_{t,m,v,f}}.\end{equation} \tag{ 2 }$$

The number of early retirements is determined by the retirement schedule, which specifies the year in which the policy takes effect (t_r , retirement year), and the minimum age of vehicles to be retired (a_r , retirement age). In the retirement year, all ICE vehicles whose ages are greater than or equal to the retirement age are forced to retire. In all subsequent years, vehicles that reach the retirement age are retired. This is shown in equation (3). In our analysis, we assume that early retirements of ICE vehicles are replaced with new sales of ZEV vehicles in the same year,

$$\begin{equation}\forall t \unicode{x2A7E} {\text{ }}{t_r},{\text{ }}f \in \left\{ {Diesel,{\text{ }}Gasoline,{\text{ }}Natural{\text{ }}Gas} \right\}:{\text{ }}{E_{t,a,v,f}} = {\text{ }}{Q_{t,a \unicode{x2A7E} {a_r},v,f}}{\text{ }}.\end{equation} \tag{ 3 }$$

Once we have determined the vehicle stock in each year of our model, we calculate the associated tailpipe emissions using equation (4), where E represents annual GHG emissions for vehicles of a given age, vehicle class, and fuel type, Q represents vehicle stock, VMT represents annual vehicle miles traveled per vehicle for a given vehicle class and age, e_f represents GHG emissions per unit of fuel for fuel type f, and fe represents fuel economy for vehicles of a given age, vehicle class, and fuel type in the specified year. GHG emissions per unit of fuel are extracted from the EMFAC database [27],

$$\begin{equation}{E_{t,a,v,f}} = {\text{ }}\frac{{{Q_{t,a,v,f}} \cdot VM{T_{a,v}} \cdot {e_f}}}{{f{e_{t,a,v,f}}}}.\end{equation} \tag{ 4 }$$

2.5.1. VMT

We include two options for calculating VMT in our model. The first is based on the age distribution of vehicles in the fleet and is unconstrained. The second is constrained by specified total VMT in each year.

2.5.1.1. Basic VMT mode

The total VMT is calculated based on the number of vehicles, the annual VMT of a new vehicle for the given vehicle class, and the VMT degradation factor, as shown in equation (5), where VMT is the VMT per vehicle for a given calendar year, model year, vehicle type and fuel type, vmt is the annual VMT by a new vehicle of type v, and d is the VMT degradation factor for model year m in year t for vehicle type v. Both the VMT degradation factors and new vehicle VMT are extracted from EMFAC [27] (see figure S14). We use this mode to calculate VMT in the BAU scenario,

$$\begin{equation}VM{T_{t,m,v,f}} = {\text{ }}vm{t_v} \cdot {d_{t,m,v}}.\end{equation} \tag{ 5 }$$

2.5.1.2. VMT profile mode

If a stock profile is specified, a corresponding VMT profile is specified (fleetwide VMT per year for each vehicle type). We use this mode to calculate VMT in the sales mandate and accelerated retirement scenarios, using the VMT profile from the BAU. This allows for a more fair comparison across scenarios. Without this constraint, VMT in scenarios with early retirements would be higher than in other scenarios due to the influx of new electric vehicles to replace the old ICE vehicles they are replacing. For HDVs, VMT is tied to the service provided, and this increase in VMT would not be expected. In this mode, the VMT is first calculated as in the basic VMT mode, and then scaled to ensure total VMT in each year for each vehicle type matches the VMT in the VMT profile, as shown in equation (6), where VMT^prof is the total VMT specified in the profile for a given calendar year and vehicle type, and VMT^adj is the scaled VMT,

$$\begin{equation}VMT_{t,m,v,f}^{{\text{adj}}} = {\text{ }}VM{T_{t,m,v,f}} \cdot \frac{{VMT_{t,v}^{{\text{prof}}}}}{{{{\mathop \sum \nolimits}}VM{T_{t,v}}}}.\end{equation} \tag{ 6 }$$

2.5.2. Fuel consumption

Fuel consumption is calculated based on VMT and fuel economy, as showed in equation (7), where FC is fuel consumption (measured in gallons of gasoline, gallons of diesel, or diesel gallon equivalents, depending on the fuel type), and FE is the fuel economy by model year and vehicle type. We assume that fuel economy remains constant over the lifetime of a vehicle (i.e. remains constant across calendar years for a given model year). We extract data on fuel consumption and VMT for each model year from CARB's EMFAC database [27] and calculate effective fuel economy (see figure S15),

$$\begin{equation}F{C_{t,m,v,f}} = {\text{ }}\frac{{VM{T_{t,m,v,f}}}}{{F{E_{m,v,f}}}} \cdot {Q_{t,m,v,f}}.\end{equation} \tag{ 7 }$$

2.5.3. Tailpipe CO₂ emissions

2.5.4. Electricity generation CO₂ emissions

In addition to emitting CO₂, both ICE vehicles and the power generation facilities producing electricity required to charge electric vehicles emit criteria pollutants including nitrogen oxides (NO_X), sulfur dioxide (SO₂), volatile organic compounds (VOCs), primary fine particulate matter (PM_2.5), and ammonia (NH₃). These species react and form secondary PM_2.5 in the atmosphere and are transported downwind. We use InMAP [35], a reduced complexity air quality model to estimate the annual average change in fine particulate matter due to emissions of these pollutants. InMAP uses a simplified representation of physical and chemical processes and transformations to map emissions of pollutants in one location to the change in annual average PM_2.5 concentration on a grid. InMAP uses a variable grid with higher resolution in more populated areas, which allows for reasonably accurate measures of exposure. As with all reduced complexity models, InMAP is less accurate than a traditional chemical transport model, but the reduced computational intensity allows for more spatial granularity and use in scenario analysis. Exposure to PM_2.5 is associated with health impacts including asthma and other respiratory diseases, cardiovascular diseases, and premature mortality [36, 37]. We estimate the mortality resulting from the increase in PM_2.5 concentrations due to tailpipe emissions and emissions associated with electric vehicle charging.

2.6.1. Statewide tailpipe emissions

2.6.2. Block-group level tailpipe emissions

2.6.3. Tailpipe impacts on PM_2.5 concentration

We use the modeled emissions to simulate the increase in PM_2.5 concentration caused by ICE vehicle criteria pollutant emissions by vehicle type. We use the InMAP Source Receptor Matrix (ISRM) [38] to map emissions in each block group to the change in PM_2.5 concentrations on a variable grid, ranging from 1 km × 1 km in densely populated areas to 48 km × 48 km in rural areas. We assume that pollutants are emitted at the centroid of each block group at ground level. We assume the distributions of VMT and vehicle population remain constant over the study period. Future work could explore the impact of changes to this distribution.

Rather than using the ISRM to calculate the change in PM_2.5 concentration caused by emissions in each year of the simulation for every scenario, which would be computationally expensive, we use the 2019 increase in PM_2.5 concentrations as a baseline, and scale the results in each year and scenario based on the annual NO_X emissions, as shown in equation (8), where PM_2.5v,s,y,x is the increase in PM_2.5 concentration in grid cell x in scenario s in year y caused by vehicle type v, and PM_{2.5v,BAU,2019,x} is the increase in PM_2.5 concentration in the same grid cell in 2019 in the BAU scenario. NO_Xv,s,y represents the statewide NO_X emissions caused by vehicle type v in scenario s in year y, and NO_Xv,BAU,2019 is the NO_X emissions caused by vehicle type v in the BAU scenario in 2019. We choose to use NO_X emissions as our scaling factor as they are the largest contributor to the increase in PM_2.5 concentrations. To test that this approach is accurate, we use the ISRM to calculate the increase in PM_2.5 for all years in the BAU scenario for one vehicle type, and compare the resulting damages to the damages using our approach. The cumulative difference in health damages over the 25 years in the study is less than 5% (see figure S16). Note that this approach would not be appropriate if the distribution of VMT and vehicle population in counties and block groups was not held constant,

$$\begin{equation}P{M_{2.5{\text{ }}v,s,y,x}} = {\text{ }}P{M_{2.5{\text{ }}v,BAU,{\text{ }}2019,{\text{ }}x}} \cdot \frac{{NO{x_{v,s,y}}}}{{NO{x_{v,BAU,{\text{ }}2019}}}}.\end{equation} \tag{ 8 }$$

2.6.4. Electricity generation impacts on PM_2.5 concentration

To determine the impacts of electricity generation required to meet the demand of electric vehicles, we use consumption-based, geographically-specific PM_2.5 concentration increase factors by balancing authority (load-serving region of the grid) from Hennessy et al [30]. Using the same approach as for tailpipe emissions from ICE vehicles, we downscale the statewide electricity consumption in each year for each vehicle type to census block groups. We then overlay the block groups with balancing authorities [39], and sum the electricity consumption in each balancing authority. We then multiply the gridded increase in PM_2.5 concentration per TWh [30] by the electricity consumption in each balancing authority.

As with tailpipe PM_2.5 impacts, we first calculate a baseline gridded increase in PM_2.5, which we use to approximate the increase in PM_2.5 in each year in each scenario. Since most vehicle types do not have electric vehicles in 2019, we do not use 2019 as the baseline here, but instead use 1 TWh of electricity consumption statewide by each vehicle type. We then scale the gridded increase in PM_2.5 by the statewide electricity consumption. As with electricity-related CO₂ emissions, we also assume a linear decrease in the increase in PM_2.5 concentration per TWh of electricity consumption as the grid decarbonizes. While more precise estimates of the change in annual emissions could be made, they would have little impact on the overall results, given the smaller scale of electricity-related emissions compared to tailpipe emissions. Equation (9) shows the approximation, where PM_2.5e_v,s,y,x is the increase in PM_2.5 in grid cell x from electricity consumption by vehicle type v, in scenario s in year y, PM_2.5e^base _v,x is the baseline increase in PM_2.5 in grid cell x from 1 TWh of electricity consumption statewide by vehicle type v, e_v,s,y is the statewide electricity consumption by vehicle type v in scenario s in year y, and p_y is the percent of 2019 PM_2.5 concentration in year y, assuming a linear decrease from 2019 to 2045,

$$\begin{equation}PM2.5{e_{{\text{ }}v,s,y,x}} = {\text{ }}PM2.5e_{v,x}^{{\text{base}}} \cdot {e_{v,s,y}} \cdot {p_y}.\end{equation} \tag{ 9 }$$

2.6.5. Premature mortality

Once we have determined the increase in PM_2.5 concentration caused by tailpipe emissions and electricity consumption in each year in each scenario, we calculate the associated premature mortality using equation (10), where M_x is the change in premature mortality, M_x ⁰ is the baseline all-cause mortality rate, β is the hazard ratio associated with exposure to an additional 10 μg m^{− 3} of PM_2.5, and P_x is the population. We use a hazard ration of 1.06 from Krewski et al [40]. We assume that the demographics and population distribution remain constant over the 25 yr period of the study,

$$\begin{equation}\Delta {M_x} = {\text{ }}M_x^0\left( {{e^{\frac{{{\text{ln}}\left( \beta \right)}}{{10}}\Delta PM2.5}} - 1} \right) \cdot {P_x}.\end{equation} \tag{ 10 }$$

2.6.6. Demographic analysis

To allow for comparison between the health and climate impacts of each scenario, we monetize both. To value the health damages, we use the VOSL, using the recommended value from the EPA [41] converted to 2020 dollars ($9.63 Million). To value the climate damages, we use the SCC in 2020 dollars. We use both the lower value previously recommended by the EPA ($51/tonne [42]) and the higher, recently adopted value of $190/tonne [43].

We use second-hand vehicle price as a proxy for the cost of retirement. We extract data from used vehicle sales databases [44, 45] for each vehicle type, and fit an exponential curve to the extracted data. We use the fitted curves to estimate the second-hand price of vehicles of each age by vehicle type. Figure S17 shows the estimated prices by age for each vehicle type. We also perform a secondary retirement cost estimation using new electric vehicle price for equivalent vehicles. For this analysis we use vehicle price data from California's Heavy-duty Vehicle Incentive Program [46].

To assess uncertainty in the results, we conduct a sensitivity analysis on three key parameters: VMT, vehicle survival, and electric grid emissions intensity. We run the model for each policy scenario while adjusting the input parameters. To assess model sensitivity to VMT we increase the VMT in each year by 10% across all vehicle types. To assess model sensitivity to survival rates, we adjust the survival profiles for each vehicle type such that the median survival increases by 5 yrs. Finally, to assess the sensitivity to grid emissions intensity, we hold the CO₂ intensity and criteria air pollutant emissions intensity constant at current levels throughout the model period. For each scenario we report the total CO₂ emissions and the total air pollution-related mortality.

Annual ICE vehicle stock and CO₂ emissions decrease significantly in scenarios with ZEV sales mandates and accelerated retirement policies. Figure 2 shows the number of ICE HDVs in the vehicle stock, and total annual CO₂ emissions from all HDVs (including electric vehicles) between 2019 and 2045 under different policies. Figures 2(A) and (C) show the stock and emissions associated with a ZEV sales mandate where the initial year of policy implementation ranges from 2025 to 2040. Figures 2(B) and (D) show the stock and emissions in accelerated retirement scenarios. Each retirement scenario shown is in addition to a ZEV sales mandate implemented in 2035. We show results for a retirement age of 5 yrs and 10 yrs, and for policy implementation ranging from 2025 to 2040.

Zoom In Zoom Out Reset image size

Under a no new policy scenario (BAU), representing current policy as of 2022, we observe a decrease in ICE HDVs from 2019 to 2045 due to fuel efficiency improvements and natural adoption of ZEVs. (see SI, figure S1). The annual CO₂ emissions under the BAU scenario decrease by 7 MTonnes from 2019 to 2045 as a consequence both of older, less efficient vehicles naturally aging out of the fleet, and moderate electrification of some vehicle types. Following the ACT regulation, which is accounted for in our BAU scenario, medium-heavy duty trucks (T6), light heavy-duty trucks (LHD1 and LHD2), and heavy-heavy duty trucks (T7, T7 OOS, and T7 Port) undergo meaningful electrification by 2045, as do buses, following the ICT regulation (see table S1 for a description of each vehicle type). The total vehicle stock is projected to increase from roughly 2 million vehicles in 2019 to 2.3 million vehicles in 2045, with roughly two thirds of the fleet being diesel vehicles (see figures S2–S11).

The implementation of ZEV sales mandates (figures 2(A) and (C)) leads to a rapid decrease of ICE vehicles and CO₂ emissions. However, even with an aggressive policy requiring 100% ZEV sales for HDVs starting in 2025 and sustained for every year thereafter, there would still be ∼354 thousand ICE HDVs on California roads in 2045, underscoring that a ZEV sales mandate alone will very likely not be sufficient for California to achieve its goal of carbon neutrality by 2045.

In contrast, the addition of an appropriate accelerated retirement policy would allow the state to meet its stated goals. Figures 2(B) and (D) show the evolution of the ICE heavy-duty stock and CO₂ emissions in selected accelerated retirement scenarios. All retirement scenarios shown are in addition to a ZEV sales mandate implemented in 2035 (the new policy in CA as of the recent Advanced Clean Fleets regulation) and assume that vehicles forced to retire early are replaced with electric vehicles. In all scenarios shown, the ICE vehicle stock and associated CO₂ emissions reach zero by 2045. We show four retirement years (the year in which the policy is implemented), and two retirement ages (the minimum age of vehicles forced to retire). Forcing younger vehicles to retire results in a quicker reduction in both ICE HDV stock and CO₂ emissions. Likewise, implementing the accelerated retirement policy sooner results in lower emissions. The ICE HDV stock decreases faster than CO₂ emissions due to emissions from electricity generation needed to meet the electricity demand from electric vehicles (see figure S12).

While a range of retirement ages were considered, 5 and 10 yrs were selected to demonstrate what would be necessary to reach a zero-emission fleet by 2045. If the state were to implement a ZEV sales mandate sooner, the minimum retirement age needed would be higher. The selected retirement ages are lower than typical retirement ages of HDVs, which range from 12 to 26 years across vehicle types based on the survival profiles used in our analysis. There is precedent for policies targeting the retirement of younger vehicles. In the nationwide Cash for Clunkers program, close to a quarter of vehicles traded in were 10 yrs old or younger [47]. However, Cash for Clunkers primarily targeted light-duty vehicles, and it is possible that targeting younger HDVs for retirement would be more challenging. This underscores the importance of beginning ZEV sales mandates sooner rather than later to reduce the need for retiring young vehicles in order to reach the state's emission goals.

While California's stated goal is reaching zero annual emissions by 2045, cumulative CO₂ emissions impact the state's overall contribution to climate change. Each decarbonization pathway results in different cumulative impacts. Figure 3 shows the cumulative CO₂ emissions from 2019 through 2045 in each policy scenario, including the BAU scenario, ZEV sales mandates beginning in various years, and accelerated retirement policies paired with ZEV sales mandates implemented in 2040 and 2035. Cumulative emissions are reduced by up to 64% in scenarios with the most ambitious policies. In the BAU scenario, 1100 MTonnes of CO₂ are emitted. 98% of these emissions come from ICE HDVs, while 2% come from electricity used in charging of the electric trucks. If a ZEV sales mandate were implemented in 2025, cumulative emissions would be reduced to 646 MTonnes of CO₂, with 11% coming from electricity generation. Removing the tailpipe emissions from ICE HDVs outweighs the added emissions from electricity generation, owing to the California grid's low carbon intensity. The earlier the ZEV sales mandate is implemented, the higher the emissions savings.

Zoom In Zoom Out Reset image size

Implementing an accelerated retirement policy would further decrease cumulative emissions. Early retirement policies on top of a ZEV sales mandate in 2035 could save up to an additional 395 MTonnes of CO₂ between 2019 and 2045 (roughly 35% of BAU cumulative emissions). Retirement policies implemented prior to a ZEV sales mandate have a greater impact on emissions than those implemented after a ZEV sales mandate.

The minimum age of vehicles forced to retire has variable impact on cumulative emissions depending on the timing of the ZEV sales mandate and retirement policy. If the ZEV sales mandate is implemented at the same time or after the retirement policy, the age of retirement has very little impact. For a ZEV sales mandate and retirement policy both implemented in 2040, the difference in cumulative emissions between retiring ICE vehicles 10 yrs and older, and retiring ICE vehicles 5 yrs and older is only 12 MTonnes. In contrast, if the retirement policy is implemented before the ZEV sales mandate, the retirement age has a larger impact. For example, for a ZEV mandate implemented in 2040 and a retirement policy implemented in 2025, the difference in cumulative emissions from retiring vehicles 5 yrs and older and retiring vehicles 10 yrs and older is 125 MTonnes, equivalent to about 1/3 of California's annual GHG emissions. This dynamic is in part due to the assumption that vehicles forced to retire are replaced with ZEVs.

These results can be explained by the fleet turnover dynamics in each scenario. When a ZEV sales mandate is implemented first, vehicles that are naturally retired are replaced with ZEVs, and there are fewer ICE vehicles remaining in the fleet when the retirement policy is implemented. In contrast, when a retirement policy is implemented first, a large number of ICE vehicles are still present in the fleet, and are forced to retire, resulting in a significant decrease in emissions due to the policy.

The total emissions in each scenario show mild sensitivity to key input parameters. Increasing VMT by 10% results in an 10% increase in CO₂ emissions across all scenarios relative to the baseline analysis. Increasing the median survival age by 5 yrs results in between a 6% decrease and a 13% increase in emissions depending on the scenario. Maintaining electricity emissions intensity at current levels results in a 3.5% increase in BAU emissions and up to 37.5% increase in emissions for policy scenarios relative to the baseline analysis due to the larger number of electric vehicles. Overall, while there is uncertainty in the magnitude of emissions in each scenario, the relative emissions of each scenario do not change, with ZEV sales mandates reducing emissions relative to the BAU, and accelerated retirement policies reducing emissions further. The sensitivity analysis suggests that the policy implications of the results are robust to changes in assumptions about electricity emissions intensity, vehicle miles traveled, and vehicle survival rates.

In addition to emitting CO₂, the use of HDVs results in an increase in fine particulate matter (PM_2.5), due to tailpipe emissions from ICE vehicles and emissions from electricity generators providing electricity to electric vehicles. As each vehicle type has different driving patterns, age distributions, and fuel type distributions, the associated air quality impacts vary by vehicle type. Figure 4 shows the increase in PM_2.5 concentration caused by emissions from each vehicle type in 2019. As we assume that the geographic distribution of VMT remains constant across all scenarios, the distribution of air quality impacts from tailpipe emissions does not change across scenarios, though the magnitude of the impacts varies. While a shift to electric vehicles results in an increase in air quality impacts near power plants, the electric vehicle damages are two orders of magnitude smaller than ICE vehicle damages. As a result, the change in distribution of impacts is negligible. Overall, the impact of HDV fleet emissions on PM_2.5 concentration is fairly low (generally less than 1 μg m⁻³ increase in all parts of the state in 2019 for each vehicle type).

Zoom In Zoom Out Reset image size

Large population centers see some of the largest impacts to air quality. The Southern portion of the state, in particular Los Angeles County and surrounding areas have the highest increases in PM_2.5 concentration across all vehicle types. The San Francisco Bay Area is impacted by medium-heavy duty trucks (T6), smaller light-heavy duty trucks (LHD1), and both local and long-haul heavy-heavy duty trucks (T7 and T7 OOS). The Northern part of the Central Valley is impacted by medium-heavy duty trucks (T6), buses, light-heavy duty trucks (LHD1 and LHD2), and heavy-heavy duty trucks (T7 and T7 OOS). The Los Angeles area and the Eastern portion of the Bay Area, where major ports are located, are impacted by heavy-heavy duty port trucks (T7 Port).

As with CO₂ emissions, cumulative air quality impacts, and the health damages caused by them, vary across scenarios. Figure 5 shows the number of cumulative deaths caused by exposure to PM_2.5 resulting from criteria air pollutant emissions from both ICE vehicles and electricity generation in selected scenarios. We do not separate the mortality from electric and ICE vehicles as the magnitude of mortality from electric vehicles is very small. From 2019 to 2045, air pollutant emissions from the BAU scenario result in over 8000 deaths. ZEV sales mandates would reduce the number of deaths by 13%–28% depending on which year the policy is implemented. The addition of an accelerated retirement policy would reduce the total number of deaths by 17%–57% for the scenarios shown.

Zoom In Zoom Out Reset image size

As with CO₂ emissions, these results are impacted by uncertainty in model inputs. Increasing VMT by 10% across scenarios results in a 10% increase in cumulative mortality for all scenarios. Increasing the median survival age by 5 yrs results in an increase in mortality between 1.1% and 21.5% relative to the baseline analysis depending on the scenario. Holding electricity emissions intensity constant at current levels increases mortality by between 0.3 and 3.0% depending on the scenario, suggesting that uncertainty in grid emissions will have little effect on the health impacts of heavy-duty decarbonization policies. As with CO₂ emissions, while the magnitude of cumulative mortality varies as the model inputs are changed, the relative mortality across scenarios remains the same, suggesting that policy implications are robust.

As shown in figure 4, the air quality impacts across the state are not uniform. Given that the population distribution is also heterogeneous, some demographic groups are impacted more than others. To assess these disparities, we combine gridded premature mortality with demographic data at the census tract level using a spatial overlay. Figure 6 shows the cumulative mortality per capita caused by each vehicle type in the BAU scenario. Figure 6(A) shows mortality per capita by household income. In general, households with low incomes experience more health damages, with the group with the highest mortality per capita being households making between $10 000 and $15 000 per year.

Zoom In Zoom Out Reset image size

Figure 6(B) shows mortality per capita by race. Across nearly all vehicle types, Latinos are the most impacted group, followed by Blacks, with the exception of buses, for which Black populations are the most impacted group, followed by Latino populations. For most vehicle types, Asian populations have the third-highest mortality per capita. The group with the lowest impacts varies by vehicle type. Long-haul HDVs (T7 OOS) impact Pacific Islander populations the least, while medium-heavy duty trucks (T6) and lighter light-heavy duty trucks (LHD1) impact Native American populations the least, and heavy-heavy duty trucks operating only within the state (T7) impact White populations the least.

For all races and incomes, long-haul heavy-duty trucks (T7 OOS) have the highest impact. This is due to these vehicles being driven more than other vehicle types as well as being less efficient and more polluting. As mentioned previously, while the magnitude of damages decreases in policy scenarios, the distribution of health damages remains the same as in the BAU. If vehicle electrification happens more quickly in some parts of the state than others, there would likely be changes in the distribution as tailpipe emissions would decrease faster in some locations than in others.

These results are driven by the geographic distribution of households and VMT. The disparities shown are likely underestimated as we are limited by our assumption that VMT is uniformly distributed across vehicles in each county, and vehicle population is distributed uniformly throughout each block group. Future work could incorporate higher resolution VMT estimates or pair this analysis with a traffic model to achieve more precise air quality and health estimates. Furthermore, our model does not track the age distribution of vehicles at the sub-state level. Older vehicles tend to be more polluting, but are driven less. While there is evidence that low-income households on average have older vehicles than high-income households [48], there is a lack of data on the relationship between income and vehicle age for HDVs. Future data collection efforts could enable an analysis of this relationship, and subsequent modeling efforts could incorporate this dynamic.

In the previous sections, we have demonstrated that accelerated retirements will be needed to allow California to meet its climate goals. However, retiring vehicles that would have otherwise continued to be used has a cost. These vehicles still have value, and to prevent them from being resold in other states without such policies, owners would need to be compensated. In this analysis, we assume that vehicle owners would be compensated by the State an amount equal to the second-hand market value of the vehicle. This assumption is representative of an incentive program in which vehicle owners trade in their older vehicles for cash towards purchasing a new ZEV. We assume that all vehicles that are retired will be scrapped and will not be resold in other states or countries.

For simplicity, we treat the cost of compensation as a constant, though there is some evidence that second-hand market value may increase as the availability of ICE vehicles decreases. The literature suggests that increases in the price of new vehicles puts upward pressure on the price of used vehicles as they become more attractive [49]. Similar dynamics could play out in the presence of ZEV sales mandates as ICE vehicles become scarce, and used ICE vehicles become more valuable. Given the uncertainty around future second-hand market prices, we perform an additional cost analysis using new electric vehicle prices to estimate the cost of retirements. This would represent compensating vehicle owners for the purchase of a replacement vehicle, rather than for the removal of their current vehicle. We use new electric vehicle prices from the California Heavy-duty Vehicle Incentive Program total cost of ownership calculator [46]. Future work could incorporate changing value of used ICE vehicles into the cost of early retirements.

To identify feasible accelerated retirement scenarios, we perform a simple cost-benefit analysis. Figure 7 shows the avoided health and climate damages vs. the cost of early retirements for each early retirement scenario presented for each vehicle type. Figure 7(A) shows the retirement costs estimated based on used vehicle price, while figure 7(B) shows the retirement costs estimated based on new electric vehicle prices. The avoided damages are calculated relative to a ZEV sales mandate alone (e.g. a retirement policy in combination with a ZEV sales mandate implemented in 2035 is compared to the ZEV mandate implemented in 2035 alone). We monetize the avoided damages using two values of the SCC and the VOSL. Points above the dashed line indicate policies that have a net benefit, where the cumulative avoided damages exceed the cumulative cost of retirements. Those below the dashed line indicate that the costs outweigh the benefits. The lower estimate uses an SCC of $51/tonne, and the higher estimate uses a value of $190/tonne.

Zoom In Zoom Out Reset image size

We first analyze the net costs using the second-hand market estimate. Using a SCC of $51/tonne (recommended by the EPA until recently [42]), most policies have a net cost. Only policies targeting heavy-heavy duty vehicles operating in ports (T7 Port) and long-haul heavy-heavy duty vehicles (T7 OOS) consistently have a net benefit. In addition, retiring long-haul heavy-heavy duty vehicles would have benefits outside of California which are not included here. For all vehicle types, the net benefit is highest for policies that are implemented sooner, and that target younger vehicles. This indicates that while retiring more vehicles is more expensive, the additional benefits of reduced health and climate impacts outweigh the additional cost.

If the EPA's current recommended SCC ($190/tonne [43]) is used, policies have a net benefit for more vehicle types. In this case, all policies forcing the early retirement of port vehicles (T7 Port), long-haul trucks (T7 OOS), heavy-heavy duty trucks (T7), smaller light-heavy duty vehicles (LHD1), and medium-heavy duty trucks (T6) result in a net benefit. Additionally, policies with earlier retirement years targeting younger vehicles result in a net benefit for larger light-heavy duty vehicles (LHD2), and buses. The vehicle category with the greatest net cost is motorhomes (MH). This is likely due to their being fairly expensive, but having low mileage each year, resulting in relatively low emissions and little benefit from removing them from the road. These results may be an underestimate of the cost of retirements given the assumption of constant used vehicle price. Incorporating the dynamics of changing used vehicle prices in response to diminishing availability of ICE vehicles could result in higher estimates of retirement cost and reduce the number of policies that are cost-effective.

Estimating the cost of early retirements using the price of new electric vehicles that would replace the retired vehicles rather than the second-hand market price of the retired vehicles results in almost none of the policies being cost effective, regardless of the SCC used. This is due to the cost of new electric vehicles being much higher than the cost of used ICE vehicles. However, this analysis assumes the price of new electric vehicles remains constant, while in reality, prices may decrease as the market grows, which could lead to additional policy options being cost-effective.