The first step of our analysis is to simulate the formation of PM_2.5, O₃, and NO₂ in the 14 study cities. To do this we perform a simulation using the chemical transport model, GEOS-Chem (Bey et al., 2001) version 35n of the adjoint (Henze et al., 2007) corresponding to version 10 of the standard forward model with additional updates to include formation processes of anthropogenic secondary organic aerosols (Nault et al., 2020). Throughout this work we will refer to the GEOS-Chem simulation of these air pollutants as the “forward model.” We conduct this simulation in the nested US domain at a resolution of 0.5° × 0.667° with 47 vertical layers. Our simulation is driven by GEOS-5 meteorological fields from the NASA Global Modeling and Assimilation Office (NASA, 2023). For all 14 cities we conduct 2-month simulations for each of the three pollutants for each month that contributes to the time-scales of the pollutant exposures we use in our health impact assessment discussed in section 2.4; overall we conduct 473 unique 2-month forward model simulations and adjoint calculations.

Our forward model configuration is thus similar to our previous work (Nawaz et al., 2021) with the following noteworthy difference. In addition to the standard NEI 2011v2.1 (US EPA O, 2015) emissions we have used previously, we now include emissions of volatile chemical products (McDonald et al., 2018) to more accurately simulate the formation of urban pollution, specifically of O₃. In doing so we better characterize emissions of anthropogenic volatile organic compounds (VOCs) that impact PM_2.5 and O₃ formation.

Once we complete the forward model simulations, we define the “response-functions” used in the adjoint calculation, for each of the 14 cities and three pollutants. For this analysis the response-functions are equivalent to “pollutant exposures” used in epidemiological studies to relate pollution exposure to health impacts and are consistent with the metric defined in the Global Burden of Disease (GBD) 2019 study (Murray et al., 2020). To calculate these exposures, we population-weight simulated pollutant concentrations, apply a mask to identify the city of interest, and, for PM_2.5 and NO₂, incorporate satellite-derived data. For the latter step, we characterize fine-scale spatial variability at a horizontal resolution of 0.01° × 0.01° by taking the ratio of the satellite-derived surface-level concentrations for PM_2.5 (van Donkelaar et al., 2021) and for NO₂ based on a previous study (Cooper et al., 2020) compared to the same satellite-derived products averaged to our model resolution. Additionally, for our PM_2.5 calculation we apply a bias correction based on the ratio of the model simulated PM_2.5 to the satellite-derived PM_2.5 for 2011. To perform the population-weighting we use the Gridded Population of the World version 4 (GPWv4) product (Center for International Earth Science Information Network - CIESIN - Columbia University, 2017). In our previous study we applied a mask corresponding to the district boundaries of Washington, DC; in this work we have applied masks derived from the Global Human Settlement (GHS) Settlement Model Grid (SMOD) (Florczyk et al., 2019) dataset consistent with recent health impact assessments (Anenberg et al., 2022; Malashock et al., 2022; Southerland et al., 2022) to define the city boundaries for each of the 14 study cities. These masks define the city extent beyond the municipal boundaries and are more representative of the metropolitan area. Lastly, we have updated the satellite-derived product for PM_2.5 from the older one (van Donkelaar et al., 2016) we used in our previous work to the most up-to-date dataset at the time of this analysis (van Donkelaar et al., 2021). This new product had an improved correlation with out-of-sample cross-validated observations (R² = 0.99) compared to the older product (R² = 0.81) based on evaluation in the study (van Donkelaar et al., 2021). We present gridded maps of the exposures at 0.01° × 0.01° resolution in Figure 1; we linearly interpolate the simulated O₃ concentrations to this resolution from the coarse model resolution. These products agree well with ground level observations as evaluated in section 3.1. The exposure values used in the adjoint calculations are provided in Figure 1.

After the exposure values (Figure 1) are calculated we conduct the adjoint calculation. During the adjoint simulation we calculate the sensitivities of these scalar exposures to emissions of their chemical precursors. For PM_2.5, we consider sensitivities to emissions of black (BC) and organic carbon (OC), ammonia (NH₃), NO_x, sulfur dioxide (SO₂), and VOCs. For O₃, we consider sensitivities to emissions of NO_x, VOCs, and carbon monoxide (CO). For NO₂ we present sensitivities to emissions of NO_x (sensitivities with respect to emissions of other species are calculated by default in the adjoint model, but are not significant, unsurprisingly, for NO₂). Corresponding to the forward model simulations, we conduct 473 2-month adjoint calculations for 2011 and aggregate the monthly calculations to get 42 unique sets of adjoint sensitivities for each of the three pollutant exposures in each of the 14 study cities. These 2-month calculations are done in parallel for each month of the year for each city; the first month is included as a spin-up period and to assess the impact of emissions from the prior month on the formation of pollutants. The adjoint characterizes the sensitivity of pollutant exposures to emissions from precursor species for each day and model grid cell. We refer to our previous work (Nawaz et al., 2021) for explicit definitions of the exposures and adjoint sensitivity calculation.

For all 14 cities, we perform several 2-month adjoint calculations to capture the influence of emissions from the previous month on the current month's air pollution exposures. In Supplementary Table S1.1 we present the percentage of the total contributions that are captured when considering days prior to the current month. Based on the results across the 14 US cities, for O₃, PM_2.5, and NO₂ we find that a 9-, 5-, and 1-day integration beyond the adjoint forcing window captures an average of 99.5, 99.7, and 99.7% of all contributions, respectively. These recommended integration windows are most appropriate for urban simulations. We note that simulations across larger spatial scales or for exposures with much lower pollutant concentrations will likely need more lag days than suggested here.

The adjoint model calculates the local-linear sensitivity of the exposures to emissions of its chemical precursors. This local-linear sensitivity characterizes the linear response of a pollutant with respect to a precursor species under the chemical conditions and emissions levels for which the forward model simulation was conducted. When applying these local-linear sensitivities to characterize changes in emissions that are notably different from this emissions magnitude, substantial uncertainty related to non-linear relationships can be incurred (Nawaz et al., 2023). To reduce the uncertainty introduced by these non-linear effects in the O₃ adjoint calculation with respect to NO_x, we apply a second-order contribution calculation derived from these local-linear sensitivities and from additional forward model simulations following the framework outlined in our previous work (Nawaz et al., 2023) for all other pollutant concentration emission relationships we only consider the local-linear sensitivities as the second-order effects are less pronounced; to elaborate, the response in NO₂ concentrations is linearly related to NO_x emissions and the response in PM_2.5 concentrations is linearly related to primary carbonaceous aerosol emissions. For PM_2.5, there are some second-order effects involved in the formation of secondary inorganic and organic aerosol; however, based on our previous analysis (Nawaz et al., 2021) these are less pronounced than the relationship between O₃ and NO_x.

Once we have calculated the adjoint sensitivities we can estimate the source contributions following the framework introduced in our previous work (Nawaz et al., 2021). In a general sense, we do so for the first order calculation as:

We calculate the exposure contribution dJ from a specific source s by multiplying the adjoint sensitivities associated with that source λ_s by the emissions associated with that source E_s. Specifically, we consider adjoint sensitivities to emissions from unique days, grid cells, and precursor species and combine them with emissions from the corresponding days, grid cells and species; we can also further disaggregate emissions by sector to estimate sectoral contributions. The second order contribution calculation is described in our previous work (Nawaz et al., 2023).

We assess the pollutant exposure impacts of proposed emission reductions in 2040 relative to 2011 by adopting an approach similar to Equation 1. In this step of the analysis, we use the adjoint sensitivities corresponding to a unique source; however, instead of combining them with emissions from a specific source, we combine them with emissions from a proposed emission scenario perturbing emissions corresponding with the grid cells, days, and precursor species of the adjoint sensitivities:

A benefit of this approach is that we can consider proposed emission scenarios (ΔE) through variations of all of the sources we identified in the source apportionment calculation. In other words, we can consider emission changes corresponding to distinct grid cells, seasons, chemical precursor species, and emissions sectors. We consider a number of emission reductions associated with proposed travel efficiency improvements described in detail in the EPA TEAM analysis (US EPA O, 2020) that specifically target emissions from on-road light-duty vehicles at the annual time-scale (Table 1). We identify the health benefits associated with areas of policy implementation using concentric circles of different radii. We iteratively increase the circle radius, centered around a city, by a fixed amount and combine the adjoint sensitivities with emissions from all of the 0.1° × 0.1° that have center points within the circle. We combine the adjoint sensitivities with emissions from larger and larger spatial areas to estimate the accrued benefits of applying these proposed scenarios to larger areas of implementation.

This approach is simple and makes two major simplifications that introduce uncertainty into our analysis: (1) we apply the percent emission reductions across all on-road vehicle emissions, including both light and heavy-duty vehicles; (2) these scenarios were developed specifically for Austin, TX so when we apply them to other cities we do not account for city-specific differences that could impact emission reductions including urban density, transit networks, sidewalks, and bike lanes. The first simplification is done because the on-road vehicle emissions used in our modeling set-up are not separated into light and heavy-duty vehicles; this likely leads to a high bias in our estimates of premature deaths avoided from the proposed scenarios. The second simplification is done because detailed EPA TEAM assessments have not been completed for most of the other study cities. We also note that the results that are presented in this work only consider the impact of the proposed travel efficiency improvement plans; we do not consider the greater reductions that have been or could be achieved from fleet turnover and current regulations, and vehicle electrification.

After calculating the contributions to pollution exposure from specific sources we assess the associated health impacts following the framework of our previous study (Nawaz et al., 2021) based on methods from the GBD 2019 study (Murray et al., 2020). For PM_2.5, the GBD 2019 study provides exposure to relative risk (RR) look-up tables which relate PM_2.5 exposure to the relative risk of premature death from six outcomes: ischemic heart disease, stroke, lung cancer, acute lower respiratory illness, type II diabetes, and chronic obstructive pulmonary disorder. We linearly interpolate between exposures and risks as they are provided at discrete intervals. For O₃ and NO₂, we apply the log-linear model used in previous works (Turner et al., 2016; Achakulwisut et al., 2019) to relate 6-month peak averaged MDA8 population-weighted O₃ and annual averaged population-weighted NO₂ to the relative risk of premature death from chronic obstructive pulmonary disorder and new pediatric asthma cases, respectively.

where X refers to the concentration above the counterfactual value and β is the concentration response factor; values are given in associated works for both pollutants (Achakulwisut et al., 2019; Murray et al., 2020). We use low-concentration thresholds of 37.6 ppbv and 2 ppbv for O₃ and NO₂, respectively, in line with our previous study (Nawaz et al., 2021).

To assess the overall health impacts of pollutant exposures we use population data at the urban level (Southerland et al., 2022) and apply age-stratifications from the state-level (Murray et al., 2020) as the relative risk exposure relationships for certain outcomes are age-dependent. For NO₂ we use pediatric populations exclusively (Anenberg et al., 2022) and do not apply this age stratification. We use disease rates from the GBD 2019 study (Murray et al., 2020) at the state level; the city boundaries of Portland, New York City, Chicago, Philadelphia, and Washington, DC extend into multiple states, for these cities we use the state where most of the population resides in, noting that bordering states generally have similar disease rates in these cities. For Washington, DC we apply state values from Virginia. We estimate the health impacts associated with some relative risk exposure relationship as:

We perform this assessment for each relative-risk health outcome and each age bracket across all three pollutants.

To characterize the health impacts associated with a specific contribution or emission reduction we perform the health impact assessment framework described above twice: once using exposures at the baseline levels (Figure 1) and a second time by multiplying this baseline estimate by the fraction attributable to the specific source sector. The latter is calculated by taking the difference between the exposure and the contribution from the source sector, divided by the total exposure value. By differencing these two values we estimate the health impacts attributable to a source or the health benefits associated with an emission reduction. We apply relative changes in baseline mortality rates, age distribution, and population data from the GBD foresight project (IHME, 2018) to account for changes in demographics by 2040. We do not project asthma incidence rates as changes in incidence rates are not considered in the GBD foresight project.

We characterize uncertainty in our exposure estimates by performing an agreement assessment between ground-level observations from the US EPA Air Quality System (AQS) and the pollutant exposures. This comparison is done for 2011 at the exposure time-scales: annual averages of daily average concentrations for PM_2.5 and NO₂ and the 6-month peak average of the average maximum daily 8-h concentrations for O₃. Observations from monitors that recorded fewer than 200 days for PM_2.5 and NO₂ and fewer than 100 days in the 6-month peak season for O₃ are excluded from our assessment. Applying this filtering, we are left with observations from 319, 1283, and 370 unique monitors for PM_2.5, O₃, and NO₂, respectively, which are located throughout the US and that cover most of the study cities. The exposures we compare against these observations are satellite-derived concentrations for PM_2.5 (van Donkelaar et al., 2021) and NO₂ (Nawaz et al., 2021), and GEOS-Chem simulated concentrations for O₃. We characterize the agreement between the exposures and observations both throughout our domain and in the study cities specifically; to do so we consider the correlation (R²), normalized mean bias (NMB) calculated as the exposure values minus observations divided by observations, and the root mean square errors (Figure 2). Biases for individual cities for all three pollutants are presented in Supplementary Table S1.2.

Performance of the satellite-derived product used for PM_2.5 (Figure 2A) is discussed in great detail at the global scale in its corresponding study. There were PM_−2.5 monitors with sufficient days of observation located within all of the study cities, shown in Figure 1, excluding Austin. Across all monitors located in our study cities the satellite-derived product agrees well with observations, with a correlation of R² = 0.63 and an NMB of −0.6%. When we compare satellite-derived concentrations in individual cities to monitors we find that NMB values range from −29.4% in Portland to +15.8% in Miami. The cities with the greatest agreement were New York and Washington, DC with NMB values of −1.3 and −2.4%, respectively. When comparing observations to the exposure values for PM_2.5 in individual cities we note that monitor sparsity could impact our assessment. Some cities only have one or two ground-level monitors; if they are located near a major source, not resolved by our exposure estimate, we would expect the exposures to be biased low. This sparsity would have an even greater effect on the shorter-lived NO₂; this is less of a consideration for O₃ given its longer atmospheric lifetime.

The only pollutant exposure that we calculate exclusively using simulated concentrations is O₃ (Figure 2B). Chemical transport model simulations of O₋₃ have been biased high (Canty et al., 2015; Travis et al., 2016) in general in the US, with among the largest high biases occurring in the south-eastern US. Due to the long atmospheric lifetime of O₃ we anticipate that less local features would be captured in our simulated concentrations than for the other, shorter-lived pollutants. With concentrations spanning a smaller dynamic range than PM_2.5 and NO₂ we also expect worse correlation. In the study cities we find a correlation of R² = 0.42 and bias of +13.2% for O₃. Unlike for PM_2.5, there were monitors with sufficient observations in all of our study cities; in these cities, biases ranged from −6.7% in Austin to +25.8% in Boston. Overall, the cities in which simulated O₃ concentrations best agreed with observations were Austin and Portland with a +7.6% bias, and Phoenix with a −8.9% bias.

Of all three pollutants, the location of ground level monitors is the most important for NO₂ due to its short-atmospheric lifetime. Although the satellite-derived approach we use to characterize fine-scale variability compares well against observations (Cooper et al., 2020, 2022; Nawaz et al., 2021) we note that this approach may be unable to capture smaller hotspots of NO₂ concentrations (associated with road systems and point sources). Beyond the inability of our approach to capture fine-scale variability in the NO₂ concentrations discussed above, the overall low bias could be attributable in part to overestimates of NO₂ from chemiluminescent monitors (Lamsal et al., 2008) that attribute concentrations of other reactive nitrate species to NO₂. Isolating our evaluation to just the study cities, excluding Austin and Seattle which had no monitors with sufficient observations, we find a correlation of R² = 0.63 and a low bias of +6.2%. There is a high degree of variability in biases for NO₂ at the city level that is likely attributable to the reasons given above: biases ranged from −63.4% in Boston to +19.4% in Portland. The cities in which the satellite-derived NO₂ concentrations had the lowest biases compared to observations were Washington, DC with a bias of +0.3%, Chicago with a bias of +5.9%, and Phoenix with a bias of +8.2%.

We next estimate the overall health impacts associated with air pollution exposure in all of our study cities (Table 2). Across these 14 cities, about 70.2 million people were exposed to air pollution in 2011; these exposures result in an estimated 24,380 PM_2.5-related deaths, 3,350 O₃-related deaths, and 126,600 NO₂-related new pediatric asthma cases in that year.

For individual cities, PM_2.5-related annual premature deaths ranged from 290 in Austin to 5,830 in New York City. Despite New York City having below average PM_2.5 as compared to the other study cities (Figure 1), demographic characteristics like population number, the age distribution of population, and baseline disease rates led to it having the highest number of PM_2.5-related premature deaths. Similar to how urban air pollution is not entirely dependent on urban emissions, air pollution health burdens are not entirely driven by exposure levels; it is important to consider differences in disease rates and population when comparing air pollution-related health impacts across cities. For example, Washington, DC and Philadelphia have similar PM_2.5 exposures and populations; however, there are 1.3 times as many PM_2.5-related premature deaths in the former due primarily to differences in baseline disease rates and age demographics. Additionally, Washington, DC and New York City have similar PM_2.5 exposures but substantially different populations which leads to 5.0 times as many PM_2.5-related premature deaths in New York City in 2011.

Premature deaths related to O₃ exposure were always substantially smaller than those from PM_2.5, consistent with previous health impact assessments (Cohen et al., 2017; Murray et al., 2020), due to a less severe relationship between exposure and relative risks (Burnett et al., 2014; Turner et al., 2016). Overall, O₃-related deaths ranged from 20 in Portland to 770 in Los Angeles. We note that low O₃ concentrations in Portland, Seattle, and to a lesser extent San Jose combined with a high degree of uncertainty in the health impact assessment could mean that the estimated O₃-related premature deaths are not significantly above zero. Comparing PM_2.5- and O₃-related premature deaths in the two most populous of our study cities, there are more PM_2.5-related deaths in New York City than Los Angeles, but the inverse is true for O₃-related premature deaths; this is attributable in part to higher O₃ exposure in Los Angeles (59.7) than New York City (54.2) despite the larger population of the latter.

For individual cities, asthma cases related to NO₂ exposure vary substantially on a city-by-city basis, more so than PM_2.5- and O₃-related premature deaths. For example, comparing the health impacts in the city with the most premature deaths (New York City) to the city with the least (Austin), there were 129 times as many NO₂-related asthma cases in New York City, but only 20 times as many PM_2.5- and O₃-related premature deaths; the latter is more consistent with the population ratio of these two cities of around 16. City-by-city variability in NO₂ exposures is greater than for PM_2.5 and O₃ exposures and areas of low and high NO₂ concentrations are generally aligned with areas of low and high population.

In the final column of Table 2, we present the anthropogenic fraction of PM_2.5- and O₃-related premature deaths in each of the study cities. This anthropogenic fraction represents the fraction of overall PM_2.5- and O₃-related premature deaths that are attributable to anthropogenic emissions. Across all cities, about 70% of PM_2.5- and O₃-related premature deaths are attributable to anthropogenic emissions, and this fraction is relatively consistent ranging from 53% in Miami to 88% in Portland. Across all of these cities a majority of air pollution-related health impacts are attributable to anthropogenic emissions. We do not include an anthropogenic fraction for NO₂-related asthma cases because, on average, anthropogenic emissions of NO_x contribute more NO₂ exposure than actual levels in the cities. This result is due to the removal of NO₂ by other chemical species, most notably reactions with hydroxyl radicals to form nitric acid, nighttime chemical reactions between NO₂ and nitrates, and the formation of peroxyacetyl nitrate and alkyl nitrates (Seinfeld and Pandis, 2016). Overall, we estimate that anthropogenic emissions of NO_x contribute on average 1.08 times the NO₂ exposure levels, ranging between 0.94 in Austin and 1.18 in New Orleans with the values >1 indicating NO₂ removal from other chemical species.

Across all of the study cities, the ratio of PM_2.5-related premature deaths to overall PM_2.5- and O₃-related premature deaths, from both PM_2.5 and O₃ exposures, was consistent. Overall, about 88% of all PM_2.5- and O₃-related premature deaths were attributable to PM_2.5 exposure ranging from 85% in New Orleans to 97% in Portland. As expected, due to the higher association between PM_2.5 exposure and premature death than O₃ exposure and premature death, PM_2.5 exposure is responsible for most of the PM_2.5- and O₃-related premature deaths in all of these cities. Lastly, considering premature deaths and asthma cases per capita allows for additional insight into the extent to which pollutant exposures and disease rates drive overall health impacts. The city with the highest per capita premature deaths was New Orleans (0.54 deaths per 1,000 people) and the city with the lowest was Seattle (0.23). Although PM_2.5 and O₃ exposures are substantially larger in New Orleans than in Seattle they are only 1.7 and 1.6 times greater indicating that baseline disease rates and non-linearities in the relative risk exposure relationships play a key role in the per capita health impacts. For NO₂, the highest per capita cases occurred in New York City (13.2 cases per 1,000 children) and the lowest per capita cases occurred in Miami (1.1).

From our results in the previous section, we conclude that the health burdens associated with exposure to PM_2.5, O₃, and NO₂ in our study cities are substantial and that a majority of these health burdens are attributable to anthropogenic emissions. Characterizing the regional influences of anthropogenic emissions on these air pollution-related health impacts is a useful step toward developing effective pollution mitigation policies. To address this, we estimate the regional contributions to urban PM_2.5- and O₃-related premature deaths and new asthma cases (Figure 3). For each city, we consider regional contributions at three levels. First, we consider the municipal contribution of the city, i.e., the extent to which emissions in a city contribute to air pollution-related health impacts in that city. Second, we consider contributions from all individual counties within the state the city is located in, excluding the municipal contribution; there is one exception, for Washington, DC we consider county-level contributions from Virginia. Lastly, we consider state-level contributions from the rest of North America, excluding prior city-level and state-level contributions. In breaking down contributions in this manner we not only characterize the extent to which emissions from a city, its state, and the rest of North America contribute to the air pollution-related health impacts in a city, but also consider the unique county-level and state-level contributions that make up these larger regional contributions.

Focusing first on PM_2.5- and O₃-related premature deaths, we consider the impact of emissions from within each city on that city; this is the “municipal” contribution. The municipal contribution varies greatly from city-to-city. For example, in Austin only 5% of PM_2.5- and O₃-related premature deaths are attributable to emissions from Austin; however, in Los Angeles and Phoenix the municipal contribution makes up over half of all PM_2.5- and O₃-related premature deaths. While a number of factors influence this variability in the municipal contribution, perhaps the most important is the city size. On average, the four smallest cities (Boston, Portland, New Orleans, and Austin) had an average area of 1,257 km² and an average municipal contribution of 20%. Contrasting this with the four largest cities (New York City, Houston, Chicago, and Los Angeles), which had an average area of 8,636 km² and an average municipal contribution of 37%, it is clear that smaller cities contribute less to their own air pollution-related health burden than larger ones. Note that the combined local and state contributions in these small cities are generally similar to the combined local and state contribution in larger cities in the same region.

The second factor that influences variability in municipal contribution is one of the most important for determining the overall breakdown of city, state, and domain-level contributions to urban PM_2.5- and O₃-related premature deaths: the geographic location of the city. In the western US, municipal, state, and domain contributions to PM_2.5- and O₃-related premature deaths make up, on average, 38, 40, and 22%, respectively, compared to the Eastern US where, on average, they make up 25, 22, and 53%, respectively. This trend makes sense given there are more cities, greater population density, and more power plants in the Eastern US. In the Western US, where urban agglomerations are further apart from one another and downwind of fewer anthropogenic sources, combined municipal and state contributions play a much greater role of 78% compared to 47% in the Eastern US. Western US states are generally larger in land area than eastern US states, which also leads to greater state contributions. By grouping cities into smaller regional clusters, e.g., Boston, New York City, Philadelphia, and Washington, DC or Austin, Houston, and New Orleans, we find that for cities in these regional clusters, contributions from emissions outside of the states along with the combined municipal and state contributions are relatively similar. The emission sectors that drive differences across these regions are discussed in detail at the end of this section.

The focus of the prior two paragraphs has been on the factors driving the regional breakdown of PM_2.5- and O₃-related premature death contributions; turning our attention to the regional breakdown of NO₂-related asthma cases, the importance of these factors is even greater. The four smallest cities (Boston, Portland, New Orleans, and Austin) have a municipal contribution to NO₂-related asthma cases of 36% compared to the four largest cities (New York City, Houston, Chicago, and Los Angeles) with a municipal contribution of 70%. Similarly, regionality is important in determining the domain contribution: in the western US the domain contribution was, on average, 5% compared to the eastern US where it was, on average 24%. The clearest difference between the regional breakdown of NO₂-related asthma cases compared to PM_2.5- and O₃-related premature deaths is the greater influence of sources near the city. As discussed previously, NO₂ has a much shorter atmospheric lifetime than PM_2.5 and O₃ so municipal emissions will have an enhanced impact on NO₂-related asthma cases compared to PM_2.5- and O₃-related premature deaths.

Another key characteristic that influences variability in the regional contributions to urban air pollution-related health impacts is the sectoral profiles of the emissions that most affect each city. We present the relative contributions from each sector regardless of location in Table 3; additionally, we present the relative sectoral contributions from the three spatial areas discussed in Figure 3 and Supplementary Tables S1.3–S1.5. We note that a majority of “other” emissions are attributable to the “surface” sector which includes a number of distinct sectors as discussed in our previous study (Nawaz et al., 2021) including non-EGU combustion, commercial cooking, gas stations, non-point industrial processing, solvents, waste disposal, and non-plume EGU and industrial emissions. County-level contributions of emissions from all sectors to PM_2.5- and O₃-related premature deaths and new asthma cases are available in the supporting information. In these county-level contributions we include lower and upper bound estimates from the health impact calculation.

Considering the results presented in Table 3, we compare the sectoral breakdowns of cities with the four largest municipal contributions of air pollution-related premature death (i.e., Los Angeles, Phoenix, Miami, and Portland) to the cities with the four smallest contributions (i.e., Austin, Boston, New Orleans, Philadelphia). Cities with more municipal contributions in general have large contributions from residential emissions (1.71 times cities with fewer municipal contributions), on-road vehicles (1.26), and non-road vehicles (1.18), but smaller contributions from power plants (0.21), agriculture (0.29), and industry (0.42). Emissions of primary carbonaceous aerosols associated with the residential sector persist for a much shorter time (Lund et al., 2018) than the precursors of secondary pollution. Subsequently, cities with proportionally more residential emissions should have a greater municipal contribution to their pollution. Emissions from on-road vehicles have been shown to have large per-mile impacts in urban areas (Choma et al., 2020, 2021) and both on-road and non-road vehicle emissions make up a large fraction of urban emissions of air pollutant precursors like NO_x and CO (Gately et al., 2017). Emission sectors associated with the longer-lived secondary components of PM_2.5 and with O₃ generally have more influence on cities with small municipal contributions; for example, emissions of SO₂ and NO_x from power plants, NH₃ from agriculture, and VOCs from industry contribute proportionally more to urban air pollution-related health impacts in cities with smaller municipal contributions.

Considering the results from Supplementary Tables S1.3–S1.5, we can characterize the relative influence of different sectors associated with different regional contributions. On-road vehicle emissions are an important contributor to urban air pollution-related premature deaths, regardless of where they are emitted; but they are especially important when emitted inside an urban boundary. On average, on-road vehicle emissions from the surrounding state and domain contribute to air pollution-related premature deaths proportionally less, 0.86 times and 0.71 times, respectively, than the municipal contribution of on-road emissions. For NO₂-related asthma cases, state and municipal contributions made up a similar percentage; however, the contribution of on-road vehicle emissions from outside of the state was proportionally 0.60 times that of municipal on-road vehicle emissions contributions. On-road vehicle emissions of NO_x from more distant locations have less influence on urban NO₂-related asthma cases than other sectors. Another sector which becomes proportionally less influential outside of state boundaries is the residential sector; municipal residential emissions make up around 10% of local contributions while only contributing 5% of contributions outside of the state.

When comparing sectoral contributions by region, a few sectors contribute a larger proportion of the outside-of-state contributions than the municipal contributions: albeit, only minorly. Considering premature deaths, outside-of-state contributions from power plants, industry, and agriculture were 3, 6, and 5%, respectively compared to 2, 5, and 1% from within cities. This also is true for NO₂-related asthma cases where outside of state contributions were 9 and 8% for power plants and industrial emissions compared to municipal contributions where these emissions made up 4 and 5%. Emissions from “other” sectors were also higher in out-of-state contributions at 47 and 45% compared to 41 and 27%; as discussed in our previous work (Nawaz et al., 2021), this other sector includes surface level emissions from power plants which could explain part of the proportionally larger contribution from other emissions outside of state.

Many actions intended to reduce urban air pollution are developed and implemented at the municipal level (Slovic et al., 2016). When developing pollution mitigation policies, local policy-makers must gauge the relative efficacy of their actions when conducted in isolation vs. in collaboration with the surrounding region. As presented in the previous section, several characteristics influence the efficacy of isolated municipal action to target air pollution-related health impacts; these include the pollutant of interest, city size, geographic location, and the sectoral sources of emissions. The framework presented in this section considers these city-specific characteristics and estimates the accrued health benefits, in the form of deaths and asthma cases avoided, that could be achieved from increasing areas implementing emission reduction policies (Table 2). In Table 4, we present the health benefits in 2040 that could be achieved by reducing transportation NO_x emissions via exclusive municipal implementations of the four proposed policies in Table 1 relative to 2011 values, from the EPA TEAM assessment; we apply the same percentage emission reductions used in Austin to all 14 study cities for emissions of primary PM_2.5, NO_x, and VOCs.

Of all scenarios considered, scenario four, a vehicles miles traveled (VMT) fee, results in the most health benefits and also corresponds to the greatest emission reductions. The implementation of this scenario would lead to 70 deaths and 1,900 asthma cases avoided in 2040 across all 14 study cities from municipal action alone. In smaller cities that have proportionally smaller municipal contributions (Austin, Boston, and New Orleans), isolated municipal action results in negligible health impact avoidances of under one premature death and under ten new asthma cases per city. A majority of the avoided premature deaths and new asthma cases occur in the two most populous cities, Los Angeles and New York City, which both have substantial municipal contributions to their air pollution-related health impacts (Figure 3). While in some cities the health benefits of municipal implementation of this VMT fee are substantial (i.e., Los Angeles and New York City), in others very few health benefits are achieved through exclusive municipal action (i.e., Austin and New Orleans).

We estimate the additional health benefits that these cities would realize from cooperative regional action by applying the proposed emission reduction percentages (Table 1) to areas of increasing size for each city (Figure 4) considering only the most effective emission reduction scenario, the vehicle miles traveled fee. The adjoint framework developed in this and our past work (Nawaz et al., 2021) requires several initial chemical transport model simulations and calculations of adjoint sensitivities for each city. These adjoint sensitivities can then be applied to many different emission scenarios, like the aforementioned areas of increasing size, to rapidly assess the impact on pollution-related health impacts from these different emission reductions without a need to perform additional simulations. With these results, we can compare the efficacy of local action relative to cooperative regional action using health impacts avoided to characterize effectiveness and identify the distances at which 50% and 75% of the total health benefits are achieved. Overall, we estimate that implementation of the vehicle miles traveled fee across the entire nested US domain, as opposed to just the municipal implementation in Table 4, would avoid 168 premature deaths and 2,771 new asthma cases across the 14 cities. By removing the municipal contributions from these totals, we estimate that the health benefits associated with non-municipal action were 1.4 times and 0.5 times that of municipal action for premature deaths and asthma cases avoided, respectively.

Across all cities, a smaller area of implementation is needed to accrue a majority of NO₂-related asthma cases avoided compared to PM_2.5- and O₃-related premature deaths. This is indicated by the sharper rise and earlier convergence of the asthma cumulative distribution functions compared to the premature death cumulative distribution functions (Figure 4). In general, cities in the western US (e.g., Los Angeles) accrue much more of their potential benefits from a smaller area of action than cities in the eastern US (e.g., New York City). In these cumulative distribution plots, we identify the distances at which 50 and 75% of benefits are achieved with square and asterisk markers, respectively. We note that smaller cities that are more dependent on regional action (Austin, Boston, and New Orleans) tend to require substantially greater regional cooperation to achieve these percentages of their potential health benefits compared to cities dominated more by municipal contributions like Los Angeles, Phoenix, and New York City.

In the center of Figure 4, we present a map of the 14 study cities along with the areas of implementation in which 75% of potential premature deaths, in magenta, and 75% of potential asthma cases, in blue, would be avoided. One of the most notable results in this figure is the distinct differences of the 75% regions for eastern and western US cities. In the eastern US, higher population density and a greater number of large cities subsequently mean that these cities would require greater cooperative action to reap the majority of benefits associated with the proposed vehicle miles traveled fee. There are two notable regions among the study cities that would require congruent action from neighboring cities in the same region to achieve 75% of potential health impacts. First, in the south-eastern US, Austin, Houston, and New Orleans would all depend on cooperative action to avoid 75% of premature deaths; additionally, Austin would require action from Houston to avoid 75% of new asthma cases; Austin and Houston are located in the same state and thus may face less roadblocks in fostering regional cooperation than other groups of cities. The second region includes the four study cities in the Northeast: Boston, New York City, Philadelphia, and Washington, DC. For each of these cities to avoid 75% of premature deaths they would require two of their neighboring cities to adopt the same proposed action and for each of these cities to avoid 75% of new asthma cases, outside of Washington, DC, they would require one of their neighboring study cities to implement the proposed action.

In general, the area of implementation needed to avoid 75% of asthma cases is much smaller than the area needed to avoid 75% of premature deaths; owing to this, exclusive municipal actions may have more success in targeting reductions in NO₂-related asthma cases than premature deaths. By comparing the sizes of these areas of implementation we characterize how capable cities are in curbing their NO₂-related health impacts compared to health impacts associated with the longer-lived PM_2.5 and O₃. In some cities, cooperation with only their own state agency is needed to achieve the 75% reduction, while in other cities, five or more state agencies need to cooperate together in order to achieve the same number of benefits within the city. Local action in New Orleans would be many times more effective in reducing NO₂-related asthma cases than PM_2.5- and O₃-related premature deaths; in this city, the area in which 75% of NO₂-related benefits are achieved is smaller than the area in which 50% of premature death benefits are achieved. Seattle is the exact opposite, the area of implementation needed to achieve 75% of the NO₂-related asthma cases avoided is nearly identical in size to the area needed to achieve 75% of the PM_2.5- and O₃-related premature deaths avoided. Ultimately, the ability of a city to address its air pollution-related health burden is highly variable, dependent primarily on the geographic location of a city and the target pollutant.