Adopting electric school buses in the United States: Health and climate benefits

Significance Whether to replace diesel school buses with cleaner electric models can be a difficult decision for local, state, and federal officials. Electric buses are expensive and their health benefits are not well known. We estimate the benefits of replacement across geographic areas to inform these decisions. When electric buses replace old diesel vehicles in large cities, health benefits associated with reduced mortality and childhood asthma total $207,200/bus. Climate benefits amount to $40,400/bus. These benefits likely exceed replacement costs. Because old buses are a large share of the current fleet, replacing them can substantially improve social welfare. Such improvements may be particularly important in low-income environmental justice communities, where budgets are often tight and health status is a major concern.


Data inputs to calculate health impacts
This section describes school bus emissions (sections 1.1 and 1.2), energy consumption (section 1.3), tire and break wear emissions (section 1.4), baseline asthma incidence (section 1.5), the calculation of attributable cases and their monetized value (section 1.6) and the mapping of county-level impacts to school districts (section 1.7).S1.Emission factors for the fleet in 2017 were calculated by Choma et al. (1) using data from the EPA's 2017 National Emissions Inventory (2).They were calculated as total school bus emissions of each pollutant in 2017 in each county divided by total school bus miles travelled in each county in 2017.We further adjust these emission factors to remove the tire and brake wear portion (section 1.4).Emission factors for MYs 2005, 2010, and 2020 are U.S. average lifetime average emission factors per mile from the GREET model calculated by Burnham (3) using EPA MOVES3.As Burnham does not provide location-specific emission factors, we apply the same U.S. average emission factors to diesel school buses MYs 2005, 2010, and 2020 driving in all locations.(2).We further adjust emission factors from Choma et al. to remove the tire and brake wear portion of PM2.5 removed (see section 1.4).Mean and 95% CI reflect the variation in emission factors by county and are weighted by school bus miles travelled.b Source: Burnham (3).For PM2.5, we use only the exhaust emission factors given by the author.For VOCs, we consider exhaust and evaporative emissions.Burnham does not provide SO2 and NH3 emission factors.

Electric school bus emissions of air pollutants and marginal mortality impacts.
Electricity grid emissions of SO2 and NOx over the bus lifetime and associated mortality impact are shown in Table S2.5)).b Calculated by Choma et al. (5).c Average of projected U.S. grid-average emissions in U.S. EIA's reference-case (6), incorporating grid losses of 4.8%.These averages weight emissions in future years discounting at 3%/year (unweighted averages are 0.171 lb./MWh for SO2 and 0.210 lb./MWh for NOx).The year of 2036 is given half weight to reflect an electric bus lifetime of 13.5 years.d Calculated as attributable deaths in 2018 x (EF2023-2036/EF2018), where EFy is the emission factor in period y.

School bus energy consumption.
We estimate fleet average diesel and electric school bus energy consumption and battery sizes using data for school buses of types A, C, and D from Levinson et al. (7), and the share of each type in the current school bus fleet using data from Lazer et al. (8) (Table S3).For electric school buses, we further assume grid losses of 4.8%, the average grid losses in 2018 in EPA's eGRID data (4), and charging losses of 10% (9).We only use diesel school bus fuel economy to calculate greenhouse-gas emissions.We do not use it to calculate diesel school bus emissions of other air pollutants.Source: Calculated from number of school buses of each type given by Lazer et al. (8).We restrict the fleet to these three most common types, assigning California's buses of types 1 and 2 given by the authors to Type A. b Source: Levinson et al. (7) for types A, C, and D. Fleet average calculated as weighted average of type A, C, and D, weighted by their share of the fleet.DSB: Diesel School Bus.ESB: Electric School Bus.
1.4.Tire and brake wear emissions.We do not include PM2.5 emissions from tire and brake wear (TBW), assuming they are similar in diesel school buses and electric school buses.For current school buses, the EPA estimates that 80% of the PM2.5 TBW emissions is due to brake wear, with tire wear making up the remaining 20% (10,11).U.S. EPA MOVES3 applies the same tire wear emissions to all vehicles of the same class (e.g., school buses) regardless of fuel type (e.g., diesel or electric) (11).For brake wear, the U.S. EPA (11) does not provide different emission factors for electric and diesel school buses but suggests that electric vehicles should have lower emissions due to regenerative braking.In this case, we might be underestimating electric school bus benefits if they reduce TBW emissions; however, this would likely be only a small underestimate since TBW emission factors are relatively small.
The emission factors we previously calculated (1) using data from the 2017 National Emissions Inventory (2) include all emissions during vehicle operation, so that we adjust them by subtracting the TBW portion from the total PM2.5 emissions.We use estimates of school bus TBW emissions per mile from EPA MOVES2014 -the MOVES version used in 2017 National Emissions Inventory -which are 0.0132 g/mi of brake wear and 0.0027 g/mi of tire wear, or 0.0159 g/mi of TBW in total (10).GREET emission factors from Burnham (3) are provided separately for exhaust and tire and brake wear, so we include exhaust only.
Although we do not include TBW emissions in our results under the assumption that they are the same in diesel and electric school buses, we estimate that the TBW portion would cause between $1,100 and $1,500 per bus in health impacts, using U.S. EPA MOVES 2014 and MOVES3 TBW emission factors (10,11) and assuming that TBW particles are equally toxic as the ambient mix by mass (Table S4).This is 30 to 50 times lower than the health impacts per bus for the average diesel school bus in the fleet in 2017 ($45,800) and almost 100 times lower relative to a MY 2005 diesel school bus (Table S5).For new MY 2020 diesel school buses, the TBW portion represents a majority of the PM2.5 emitted and causes more impact than exhaust PM2.5, since MY 2020 emit relatively low amounts of exhaust PM2.5.However, when considering all pollutants, MY 2020 diesel buses cause $7,600 in health impacts (a large majority from NOx), a figure that is still 5 to 7 times larger than the impacts of TBW emissions.Therefore, even if electric buses were to cause small increases in TBW emissions, this would have a very small impact in our results, even for new MY 2020 buses.Source: Sum of tire and brake wear emission factors given by U.S. EPA (10).b Source: Sum of tire and brake wear emission factors given by U.S. EPA (11).(2), with the tire and brake wear portion of PM2.5 removed.b Source: Emission Factors from Burnham (3).

Baseline Asthma Incidence.
To calculate baseline asthma incidence, we apply nationallevel age-specific incidence rates from Winer et al. (12) to the population at risk in each age group in each county in the contiguous U.S. (Table S6).Winer et al. estimate 12-month incidence rates using data from 2006 to 2008.Data for 2006 covered 24 states and DC, whereas data for the other two years covered 34 states and DC.We define the population at risk as children without asthma.We determine the population at risk in each county and age group using singleyear-of-age population counts from the National Vital Statistics System (13), subtracting the proportion of children estimated to have asthma in 2019, using national-level prevalence rates from the 2019 National Health Interview Survey (NHIS) (14).(14).For children older than 4 years, the 2019 NHIS give prevalence rates of 9.1% for ages 5 to 14 years and 7.4% for ages 15 to 19 years.We apply the 9.1% rate to the 5 to 11 year-old-group in the present study, and the average rate between those two NHIS age groups (8.25%) to the 12 to 17 year-old-group in the present study.c Calculated as 1 -Asthma Prevalence.We then determine the number of children at risk by applying these percentages to single-year-of-age population counts for each county from the National Vital Statistics System (13).
1.6.Calculating attributable cases and monetized health impacts.Marginal impacts per mass emitted of each pollutant are assessed with Eq.S1, following Choma et al. (1), which yields attributable new asthma cases.

Marginal Impacts
Where (1): M is the baseline outcome measure (new childhood asthma cases per year); AM is the attributable outcome measure (new childhood asthma cases per year); C is the ambient PM2.5 concentration; DC is the change in ambient PM2.5 concentration; and RR is the relative risk from the concentration-response function.The index p = 1,2,…,5 represents different pollutants, index a represents different age groups, and indices s and r represent source and receptor cells (52,411 InMAP cells (15,16), which were mapped to 3,108 counties in (1)).In our analyses of impacts occurring in certain regions (e.g., inside vs. outside each metropolitan area), we sum receptors accordingly (e.g., r inside the metropolitan area of s, and r outside the metropolitan area of s).
Following Choma et al. (1), we implement this calculation computationally using Eq.S2, which incorporates the mapping from ISRM cells to counties.Eq.S2 yields monetized impacts as we multiply attributable cases by a value per statistical case (VSC) of childhood asthma.
Where: MI is a matrix where MIi,j is the monetized value of the marginal impact on new childhood asthma cases occurring in county j=1,2,…,3108 (receptor) as a consequence of 1 metric ton of emissions in county i=1,2,…,3108 (source); P is a matrix where Pij is the percentage of the population of county i=1,2,…,3108 that is within InMAP cell j=1,2,…,52411; ISRM is a matrix where ISRMij is the increase in concentration ΔC [μg/m 3 ] in InMAP cell j=1,2,…,52411 (receptor) as a consequence of emissions of 1 μg/s in InMAP cell i=1,2,…,52411 (source); M is a diagonal matrix where Mij is attributable new childhood asthma cases for an increase in 1 μg/m 3 in ambient concentration in county i=j=1,2,…,3108 if i = j and 0 if i ≠ j; k = 10 12 /(24 × 3,600 × 365) represents the conversion from 1 μg/s to 1 metric ton/year; and VSC=610,000 [2022 USD] is the value per statistical case (17).The index p=1,2,…,5 represents the pollutants.M is calculated as M = Dg(diag(D × S T )), in which D is a matrix where Dij is the number of deaths in county i=1,2,…,3108 and age group j=1,2,3; S is a matrix where Sij is the percent increase in baseline mortality in county i=1,2,…,3108 and age group j=1,2,3 for an increase in ambient concentrations of 1 μg/m 3 ; diag(X) denotes the vector containing the diagonal elements of matrix X; and Dg(x) denotes the square matrix where off-diagonal elements are 0 and diagonal elements are the elements of vector x.
We account for mortality impacts using our previous estimates of mortality per mass emitted of each pollutant in each county (1,18), which were previously calculated using the same method, using different input data necessary to calculate mortality risks: baseline mortality, relative risks from a concentration-response linking exposure to ambient PM2.5 to mortality risks, and a value per statistical life (VSL) (as opposed to the childhood asthma VSC).We further adjust these previously calculated impacts to 2022 dollars and income levels by adjusting the VSL (see main manuscript, methods).
In our previous study (1), we developed estimates of attributable mortality and their economic value in 2017 using different concentration-response functions.In the current study, we use the version that used the age-specific concentration-response functions from the Global Exposure Mortality Model (GEMM) (19), which was our base case in Choma et al. (1).Many cohorts studying the relationship between ambient PM2.5 exposure and mortality have been conducted to date and GEMM is an important synthesis of this evidence.GEMM was a collaboration among 15 research groups responsible for 15 of the largest cohorts, where the investigators fit the GEMM concentration-response function directly to individual-level data (19).In addition, it also incorporated published effect estimates for another 26 cohorts.GEMM provides different concentration-response functions, and we used version that applies to age-specific nonaccidental mortality, and that includes a recent Chinese Male Cohort.GEMM's concentrationresponse function is nonlinear in ambient PM2.5 levels, with steeper slopes at low concentrations.Although we use the age-specific GEMM concentration-response functions, whose slopes vary, the GEMM version that applies to all adult non-accidental mortality yields a slope of 1.05% increase in adult non-accidental mortality per each 1 µg/m 3 increase in ambient PM2.5 exposure at the mean U.S. ambient PM2.5 concentration.

Estimating impacts by school district.
To help inform policies focusing on specific school districts, we produce estimates of health impacts benefits for each 13,309 local education agencies (school districts) in the contiguous U.S. (20,21) by taking the school-age-populationweighted average of the county-level impacts.We use (i) the percentage of each school district intersecting each Census Block Group, using 2023 relationship files for local education agencies (school districts) from the National Center of Education Statistics (20); and (ii) the population aged between 5 and 19 years old for each Census Block group, using 5-year estimates from the 2022 American Community Survey from the U.S. Census Bureau (22), which were then aggregated to counties.For the state of Connecticut, both the NCES relationship files and the 2022 ACS population estimates use the new 2022 Census block groups, following the new 2022 county equivalents.In this case, we map these to the pre-2022 county equivalents in -used in our air pollution model -using 2022 county subdivision to 2020 block group relationship files for Connecticut from the U.S. Census Bureau (23) and total population in each Census block group in 2020 from the Decennial Census, from the U.S. Census Bureau (24).

Supplementary Results
Figures S1-S11 show our supplementary results.

1. 1 .
Diesel school bus emissions of air pollutants.Emission factors for the fleet in circulation in 2017, as well as for diesel school buses of model years (MYs) 2005, 2010, and 2010 are shown in Table

Figure S1 .
Figure S1.Health impacts of school buses: PM2.5-attributable deaths per bus, by bus model year, driving location, pollutant species, and outcome.Locations are classified using NCHS's Urban-Rural classifications(25).DSB: Diesel School Bus.ESB: Electric School Bus.

NH 3 MarginalFigure S6 .
Figure S6.Distributions of marginal damages, including only mortality attributable to PM2.5, per ground-level emissions of 1 metric ton of each species in each county.Source: adjusted from Choma et al. (1), adjusting the value per statistical life to 2022 dollars and income levels.

Primary PM 2 . 5 Marginal
Damage [2022 USD per metric ton], log scale Number of Counties Distribution of marginal damages for each species by county −− Mortality [unweighted, for the 3,108 counties in the contiguous U.S.

Figure S8 .Figure S9 .
Figure S8.Cumulative distributions of marginal damages, including both mortality and new childhood asthma cases attributable to PM2.5, per ground-level emissions of 1 metric ton of each species in each county.Cumulative distributions weighted by total school bus emissions of each species and each county in 2017.Mortality marginal damages are adjusted from Choma et al.(1), adjusted to reflect the value per statistical life in 2022 dollars and 2022 income levels.

Table S1 .
Real-world diesel school bus emission factors per mile for each pollutant, by school bus model year.
a Source: Calculated by Choma et al. (1) using 2017 National Emissions Inventory data from EPA

Table S2 .
Marginal impacts from electricity grid emissions.

Table S3 .
Diesel and electric school bus energy consumption.

Table S4 .
Tire and Brake Wear emission factors and associated health impacts.

Table S5 .
Exhaust PM2.5 emission factors and diesel school bus health impacts.

Table S6 .
Parameters used to calculate baseline asthma incidence.Age group [years] Asthma Incidence a Asthma Prevalence b [%] Population at risk c [%] a Source: Winer et al. (12) b Source: 2019 National Health Interview Survey

PM 2.5 −attributable new childhood asthma cases per bus
(25)re S3.Health impacts of school buses: PM2.5-attributable deaths per mile driven, by bus model year, driving location, pollutant species, and outcome.Locations are classified using NCHS's Urban-Rural classifications(25).DSB: Diesel School Bus.DSB:

PM 2.5 −attributable new childhood asthma cases per mile
Figure S5.Distributions of marginal damages, including mortality and new childhood asthma cases attributable to PM2.5, per ground-level emissions of 1 metric ton of each species in each county.Primary PM

.25M VOC Marginal Damage [2022 USD per metric ton], log scale Number of Counties Distribution of marginal damages for each species by county −− Mortality [unweighted, for the 3,108 counties in the contiguous U.S.]
] Distributions of marginal damages, including only new childhood asthma cases attributable to PM2.5, per ground-level emissions of 1 metric ton of each species in each county.

Cumulative Health Impact Per Year [2022 USD] Figure
(25) Per-vehicle health impacts and benefits of school bus electrification, by bus model year, driving location, pollutant species, and outcome.Locations are classified using NCHS's Urban-Rural classifications(25).DSB: Diesel School Bus.ESB: Electric School Bus.DSB:

Fleet in 2017 DSB: MY 2005 DSB: MY 2010 DSB: MY 2020 2022 USD per bus
(25)vehicle health impacts and benefits of school bus electrification, by bus model year, driving location, pollutant species, and outcome: sensitivity analysis assuming carbonaceous particles are five times more toxic than the ambient mix, by mass.Locations are classified using NCHS's Urban-Rural classifications(25).DSB: Diesel School Bus.ESB: Electric School Bus. DSB: