Criteria pollutant emissions were predicted with the California Regional Multisector Air Quality Emissions (CA-REMARQUE) model (Zapata et al., 2017) for the BAU and GHG-Step scenarios. Both scenarios were constructed using CA-TIMES, a technology-rich, bottom-up, energy economics model that determines the least-cost mix of technology–fuel options for all sectors of the statewide economy. CA-REMARQUE translated these behavior, technology, and fuel changes into spatially and temporally resolved criteria pollutant emissions inventories. CA-REMARQUE predicted that adoption of the GHG-Step policies in place of the BAU policies would cause decreases in emissions of primary PM0.1 (-36 %), PM2.5 (-3.6 %), oxides of nitrogen (NOx, -13.7 %), and ammonia (NH3, -27.5 %) but cause increases in emissions of carbon monoxide (CO, +37 %) and oxides of sulfur (SOx, +14 %). Some components of primary PM2.5 emissions responded more strongly to different technology changes yielding nonuniform reductions of PM2.5 elemental carbon (elemental carbon, -10.6 %), PM2.5 organic carbon (organic carbon, -13.3 %), and PM2.5 copper (Cu, -63 %). The spatial allocation of emission rates was determined by either using existing 4 km spatial patterns of emissions sources or finding new optimal locations for new emissions sources such as biorefineries that were placed near high-biomass-feedstock regions. The future BAU and GHG-Step scenarios considered in the present study do not include nuclear or coal-fired (with or without carbon capture and sequestration) electricity generation in California. Electricity generation in the 2050 GHG-Step scenario is dominated by wind (34 %), solar (34 %), and natural gas (18 %) with smaller contributions from tidal, geothermal, and hydro. A comprehensive analysis of all emissions changes including spatial plots is provided by Zapata et al. (2017).

Meteorology simulations using the Weather Research and Forecasting (WRF) model v3.2.1 (University Corporation of Atmospheric Research 2010) conducted previously (Zhang et al., 2014) for the years 2048–2054 were used as meteorological inputs in this study. Hourly-averaged fields describing spatial and temporal wind speed and direction, humidity, temperature, planetary boundary layer height, downward shortwave radiation, air density, and precipitation were formatted for use with the regional chemical transport model. The 2054 calendar year was the median year over the period 2048–2054 for domain-average PM2.5 concentrations within the South Coast Air Basin that contains the majority of the population in California. The focus of the current study is to evaluate how the emissions changes lead to different air quality outcomes. Both emissions scenarios are evaluated using the same meteorology, which minimizes the variability introduced by the climate signal.

Air quality was simulated using the UCD-CIT (University of California, Davis – California Institute of Technology) 3-D regional chemical transport model (Kleeman and Cass, 2001; Ying et al., 2007; Hu et al., 2015). The SAPRC11 (Carter and Heo, 2012; Carter et al., 2012) chemical mechanism was used to represent gas-phase chemical reactions. Gas-to-particle conversion was simulated as a dynamic process based on the concentration of semi-volatile gas-phase compounds at the particle surface in equilibrium with the condensed material inside each particle. Thermodynamic equilibrium within each particle for inorganic species was calculated using the ISORROPIA model (Nenes et al., 1998). Thermodynamic equilibrium within each particle for organic species was calculated using a two-product model (Carlton et al., 2010). Particulate matter emissions profiles include 18 organic, inorganic, and metal particulate species distributed across 15 size bins.

Air quality simulations were conducted over three horizontal domains, a coarse 24 km parent domain, and two 4 km resolution child domains. The coarse domain covered all of California and the adjacent Pacific Ocean to provide boundary inputs to the higher-resolution child domains over populated regions in northern and southern California. A total of 16 telescoping vertical layers were used up to a total height of 5 km above ground. Simulations were conducted for the first 28–29 days of each month for the 2054 calendar year. The first 3 days of every month were excluded to minimize the effects of initial conditions which are not known exactly, leaving 301 simulation days to be used in the statistical analysis.

A 2050 California population projection at 4 km spatial resolution was used for both population-weighted concentration estimates and mortality estimates. This population projection is based on the highly resolved block-group 2010 census population data in shapefile format (US Census Bureau) which was intersected with the regular air quality grid. The 4 km resolution population field was then scaled according to the projected populations for each county in 2050 (California Department of Finance. Demographic Research Unit 2014) relative to 2010 (Table S2). This procedure was conducted separately for population age > 35 and for all ages (see Fig. S1) to be used for the population-weighted code (all ages) and the mortality estimates (> 35 years). The combined southern and northern 4 km resolution modeling domains encompassed 92 % of California's projected 2050 population (summarized in Table S3).

Population acts as a spatial surrogate for distributing emissions and as a receptor for calculating the public health effects of air pollution. Consistent population fields were used for both of these tasks in the current study. Population growth rates by county are summarized by Zapata et al. (2017).

Several statistical analyses were conducted across space, seasons, and scenarios. Annual-average concentration plots were estimated by taking the average of 301 daily concentration fields. A two-tailed paired t test was used to identify significant differences between BAU and GHG-Step concentrations. Annual or seasonal concentration field plots were condensed to a statewide, air basin, or county population-weighted concentration estimate by summing the concentration multiplied by the population in each cell and then dividing the resulting sum by the entire population for the region of interest.

Daily maximum 8 h average O3 concentrations were calculated for each model grid cell. Subsequent seasonal or annual averages used the daily maximum 8 h average concentrations for a given state, basin, or county. To determine whether a county was in compliance with the 70 ppb O3 National Ambient Air Quality Standards (NAAQS), the fourth highest population-weighted maximum 8 h O3 concentration was calculated. The number of days exceeding this standard was also tabulated.

Premature mortality estimates from long-term exposure to PM2.5 and O3 were calculated using annual-average 4 km resolution concentration fields for the BAU and GHG-Step scenarios. The attributable fraction (AF) is the portion of deaths or incidences that can be associated with the cause of interest, in this case the fraction of deaths due to annual PM2.5 and O3 exposure. The AF quantifies the change in the relative risk. AFi=RRi-1RRi=eβ(xi-xi,bkg)-1eβ(xi-xi,bkg)

The log-linear incidence rate function is assumed when calculating the risk ratio (RR) as shown in Eq. (1). The beta coefficient (β) is derived from taking the natural log of the RR found in epidemiology literature. PM2.5 RR for all-cause mortality associated with a 10 µm m-3 increase in long-term PM2.5 exposure is estimated at 1.062 based on a worldwide meta-analysis (Hoek et al., 2013) or 1.036 based on the American Cancer Society follow-up (Krewski 2009). An O3 RR of 1.04 for respiratory mortality from long-term O3 exposure is based on Jerrett et al. (2009). The change in concentration is based on taking the annual-average concentration for a given grid cell (xi) and subtracting it from the background concentration (xi,bkg). Background concentrations on the west coast of North America are often measured at mountain sites that sample the free troposphere. Herner et al. (2005) measured PM1.8 concentrations of 4 µg m-3 at Sequoia National Park (elevation 535 m) during periods when this site was in the free troposphere. McKendry (2006) surveyed published literature and reviewed monitoring data in British Columbia on the west coast of North America and estimated that background PM2.5 concentrations are 2 µm m-3 with little evidence of change over time. A background PM2.5 concentration of 3 µm m-3 and O3 concentration of 35 ppb was assumed in the current study. The beta coefficient, change in cell concentration, is then used to calculate the risk ratio (RRi) and subsequently the AF.

Es=∑iAFiBcPi

The mortality (Es) for each scenario for a given region was calculated using Eq. (2) by taking the product of the population and mortality rate to get the deaths, followed by multiplying the fraction that is attributable to pollution (see Eq. 1). Population (Pi) projections for ages 35 and older were used in this calculation due to high uncertainty for younger age groups. Averaged 2009–2013 California all-cause (all ICD 10 codes) and respiratory (ICD 10 codes J0–J98) mortality rates (Bc), calculated in deaths per 100 000, were determined for each California county for ages 35 and older from the CDC WONDER database (United States Department of Health and Human Services (US DHHS) et al., 2014).

Costs associated with premature death from long-term air pollution exposure were estimated using the VSL method, assuming that a death equates to USD 7.6 million, based on the distribution of 26 economic reports (Viscusi and Aldy 2003) and the suggested value by the EPA (Industrial Economics, 2011; Bart Ostro, personal communication, 2015; RTI International 2015). This value can be adjusted to a future year with an average discount rate “i” by multiplying with the value (1+i)futureyear-baseyear, where the base year is 2006. VSL is estimated based on a willingness to pay for small reductions in mortality risk through the selection of different job types. “Willingness to pay” estimates are thought to incorporate “cost of illness” including morbidity but they do not capture non-health damage.

PM2.5 concentrations can be analyzed on timescales ranging from seconds to years, but annual-average PM2.5 concentrations are most commonly used to calculate mortality and health damages. Figure 5 illustrates annual-average PM2.5 concentrations in northern-central California and southern California in 2054 under the BAU scenario (Fig. 5a) and the differences induced by the GHG-Step scenario (Fig. 5b). Both results use identical 2054 meteorology, ensuring that the concentration differences reflect changes between each scenario's emissions inventory. The highest BAU annual-average PM2.5 concentration in southern California is ∼ 18 µg m-3 in the city of San Bernardino located east of Los Angeles, with the next highest PM2.5 hot spots occurring at San Diego and near the busy Port of Los Angeles and Long Beach. In Northern California, the annual-average PM2.5 peaks at 25.3 µg m-3 between the cities of Oakland and SF. Maximum PM2.5 reductions in the GHG-Step scenario (Fig. 5b) occur between Oakland and SF (-6 µg m-3), in SD county (-5.3 µg m-3), and in San Bernardino county (-3.5 µg m-3). Overall, the reductions are significant (p value ≤ 0.1) over the majority of northern and southern California; the only non-significant PM2.5 changes are two locations inland in northern Los Angeles around Lancaster and in midwestern San Bernardino, where BAU concentrations were low. Significant PM2.5 increases of +0.5 µg m-3 do occur in ocean shipping routes because more fossil fuel is used for marine vessels in the GHG-Step scenario than in the BAU scenario. The GHG-Step scenario requires increased biofuel use as part of the overall strategy to reduce GHG emissions. This increased biofuel production is associated with higher biofuel costs since the least expensive biofuel feedstocks are used first followed by progressively more expensive feedstocks. As biofuel utilization increases, the demand and cost for conventional fossil fuels decreases. The decreased cost for fossil fuels in the GHG-Step scenario makes these fuels attractive for use by marine sources.

Population-weighted PM2.5 concentrations (Fig. 2b) decrease for all regions in all seasons under the 2050 GHG-Step scenario relative to the BAU scenario. Variability in PM2.5 concentrations is highest during the winter, with periods of intense stagnation intermixed with periods of vigorous atmospheric mixing. PM2.5 concentrations are less variable in the summer months as demonstrated by the smaller IQR in Fig. 2b. The annual population-weighted PM2.5 concentration drops from 6.0 to 4.8 µg m-3 (-20 %) in the SJV, 8.3 to 6.2 µg m-3 (-25 %) in SD, 9.5 to 7.8 µg m-3 (-18 %) in SF, and 7.6 to 6.5 µg m-3 (-14 %) for the SC air basin. Additional detail of the PM2.5 species that decreases the most (e.g., nitrate) and the changes in the particulate size distribution are further described in the Supplement and summarized in Table S5.

Certain PM2.5 spatial patterns illustrated in Fig. 5 were difficult to anticipate based exclusively on statewide emissions totals. For example, the PM2.5 co-benefits from widespread adoption of new vehicle technology contribute significantly to statewide emissions reductions, but these changes were distributed over a larger area than the benefits associated with the decarbonization of freight modes (e.g., rail, aviation, and marine). Most on-road vehicles in California already have relatively low emissions rates for criteria pollutants. Further vehicular emissions savings result from small reductions that are distributed over the large number of vehicles across the entire state. This spreads the air quality improvements associated with vehicles over a large area. In contrast, freight modes use fuel with higher sulfur content burned in engines with less aftertreatment control (e.g., particulate filter) leading to higher particulate matter emission rates per energy consumed (e.g., mg J-1). These sources are localized to goods movement corridors (shipping lanes, rail lines, etc.) that intersect at transport distribution hubs near ports. This leads to localized reductions in particulate matter concentrations associated with freight modes compared to more diffuse reductions associated with on-road sources. These trends were not obvious from statewide emissions tables but are clearly illustrated by the results from regional air quality modeling.

Figures 6 and 7 illustrate the deaths, death rate, and cost associated with premature deaths from long-term annual exposure to both PM2.5 and ozone (O3).

Using a VSL equal to USD 7.6 million per avoided death (Industrial Economics, 2011; Bart Ostro, personal communication, 2015), total costs for premature deaths in California equal ∼ USD 47.0–78.5 billion yr-1 in the 2050 BAU emissions scenario, with a savings of USD 11.4–20.4 billion yr-1 in the GHG-Step emissions scenario (right axis Fig. 7a). Los Angeles County has the highest premature mortality associated with air pollution (25 % of California) and thus the highest air pollution mortality cost under all emissions scenarios. Air pollution damages in Los Angeles County are valued at USD 15.2–25.5 billion yr-1 in 2010, which decreases to USD 12.1–19.6 billion yr-1 in 2050 BAU. Adoption of the GHG mitigation strategies in California reduces air pollution damages in Los Angeles County by USD 1.9–3.6 billion yr-1 (17–18 % reduction). Other major counties also experience reduced air pollution costs under the GHG-Step scenario relative to BAU, including SD (USD 1.7–2.9 billion yr-1 reduction; 15 %–16 %) and Sacramento (USD 0.70–1.3 billion yr-1 reduction; 6.4 %). However, the largest cost savings per capita are predicted to occur in and around counties near SF based on the higher mortality rate reductions.

The costs for reducing California GHG emissions 80 % below 1990 levels by the year 2050 depend strongly on numerous assumptions about external factors such as the global price of oil. Only a few California energy models are available that attempt to calculate costs across the entire economy (Morrison et al., 2014, 2015). Analyses produced by the E3 PATHWAYS model (Williams et al., 2012; Energy+Environmental Economics (E3), 2015) suggest that meeting an intermediate target (40 % reduction in GHG emissions by the year 2030) using a non-optimized energy portfolio scenario would reduce personal income by USD 4.95 billion yr-1 (-0.15 %) and lower overall state gross domestic product by USD 16.1 billion yr-1 (-0.45 %). An analysis produced by the CA-TIMES model (Yang et al., 2014, 2015) indicates that the optimized GHG-Step scenario is less expensive than the BAU scenarios.

The air pollution analysis carried out in the current study predicts that the GHG-Step scenario will provide public health benefits equivalent to USD 11.4–20.4 billion yr-1 relative to the BAU scenario in 2050. The public health benefits described here have relatively tight uncertainty ranges with median values that are comparable to the more pessimistic of these two cost estimates for the adoption of low-carbon energy.

Figure 8 illustrates the public health savings associated with the GHG-Step scenario alongside the “fair-share” benefits of federal programs (United States Office of Management and Budget, 2016) that affect California. Fair-share benefits are calculated using the fraction of US residents living in California multiplied by the total US benefits. The GHG-Step scenario yields benefits that are larger than those from of any program under the Federal Department of Agriculture, Energy, Health & Human Services, Labor, and Transportation. Only the National Ambient Air Quality Standards (NAAQS) under the US EPA have greater public health savings associated with reduced concentrations of air pollution. As shown throughout Sect. 3, strategies to reduce GHG emissions have benefits that overlap with NAAQS objectives and produce air quality improvements that would otherwise be challenging or impossible to achieve under the BAU scenario.

Taken together, the immediate and long-term savings associated with the GHG-Step scenario make a compelling case for the shift to a low-carbon energy system in California.