State-level drivers of future fine particulate matter mortality in the United States

The global change assessment model (GCAM) is a partial equilibrium human-earth system model that represents the linkages between global energy, water, land, climate, and economic systems (Calvin et al 2019, JGCRI 2019a). In GCAM, exogenous population, GDP, and labor productivity assumptions drive overall energy and service demands, and a logit function (McFadden 1973, Clarke and Edmonds 1993) is used to estimate the most economically feasible and technically viable combination of technologies that satisfy these demands in each market and for each modeling period. Corresponding air pollutant and greenhouse gas emissions are also estimated. A recent version of GCAM (v5.1) has been evaluated by comparing to historical data and to future scenario simulations from other models (Calvin et al 2019). GCAM has been widely used to explore global energy and emission scenarios, including the Intergovernmental Panel on Climate Change Special Report on Emissions Scenarios (Nakicenovic et al 2000), Representative Concentration Pathways (Thomson et al 2011), shared socioeconomic pathways (SSP) (Calvin et al 2017), and two-degree mitigation scenarios (Fawcett et al 2015).

GCAM-USA is an extension of GCAM with representation of US state-level energy supply and demand markets (Zhou et al 2014, Iyer et al 2017, JGCRI 2019b). State-level population and economic growth assumptions are calibrated to historical levels through 2015. For future modeling years, population projections are derived from the downscaled shared socioeconomic pathways 2 (SSP2) 'Middle-of-the-Road' scenario (Jones and O'Neill 2016, Jiang et al 2018), being qualitatively consistent with assumptions made in the rest of the model. State-level per capita GDP growth rates (table S2 is available online at stacks.iop.org/ERL/14/124071/mmedia) are harmonized with regional assumptions in the National Energy Modeling System model used in the 2016 Annual Energy Outlook (AEO) (EIA 2019a).

In contrast to stylized business-as-usual scenarios, this study reflects major air quality and energy policies currently in place, and how these constrain state-level emissions. In previous work (Shi et al 2017), we incorporated US air pollutant emission coefficients, pollution controls for coal-fired power plants, and representations of major energy and air quality regulations, and we adjusted the base year emissions and future trends to be consistent with the 2011 National Emission Inventory and its projections (EPA 2015). Here, we adopt improved representations of building sector energy demands, electric vehicle costs, and retirement of existing coal and nuclear electric power plants, and harmonize total electricity generation and electricity generation from coal with the 2018 AEO (EIA 2019b) (SI section 1). Additionally, a nine-state regional cap on electric sector CO₂ emissions was added to reflect the regional greenhouse gas initiative (RGGI) (RGGI 2019).

We project future changes in PMMC based on monetized PM_2.5-related mortality impacts using estimates of willingness-to-pay to avoid these outcomes. This method has been used by US EPA in benefit-cost analyses of the Clean Air Act (EPA 2011b). Mortality impacts account for over 90% of the total monetized PM_2.5 health impact (Martenies et al 2015). PM_2.5, in turn, dominates total air pollutant mortality burdens (Zhang et al 2018).

Our initial application of GCAM-USA to model PMMC (Ou et al 2018) utilized national average year-, pollutant-, and sector-specific PMMC coefficients (Fann et al 2012). Projections of these coefficients to 2050 accounted for national changes in population and economic growth. This paper improves upon Ou et al (2018) by applying PMMC coefficients that differ by state, accounting for the spatial heterogeneity of pollutant transport and chemistry, population, and baseline mortality rates. The reduced-complexity model EASIUR (Heo et al 2016a), derived from tagged CTM simulations, has estimated all-cause mortality costs of PM_2.5 ($/tonne) due to air pollutant emissions (primary PM_2.5, SO₂, NO_x, and NH₃) from each county in the US in 2005, using the concentration-response function (equation (1)) from the Krewski et al (2009) with a relative risk (RR) of 1.06 (95% confidence interval, CI of 1.04–1.08) and a value of statistical life (VSL) of $8.6 million (2010 USD)

$$\begin{eqnarray}&&{\rm{\Delta }}{\rm{P}}{\rm{M}}{\rm{M}}{\rm{C}}={y}_{0}\cdot \left\{1-\exp \left(-\displaystyle \frac{\mathrm{ln}\,{\rm{R}}{\rm{R}}}{10}\cdot {\rm{\Delta }}c\right)\right\}\cdot {\rm{V}}{\rm{S}}{\rm{L}},\end{eqnarray} \tag{ 1 }$$

where ${y}_{0},$ the baseline mortality, is the product of baseline mortality rate and population for given grid, ${\rm{\Delta }}c$ is the change in PM_2.5 concentration, and RR is the relative risk, representing the change in mortality rate for an increase of 10 $\mu {\rm{g}}\,{{\rm{m}}}^{-3}$ in PM_2.5 concentration.

In our work, these county-level per-tonne mortality cost estimates are aggregated to the state level for the electricity, industry, transportation, and building sectors in 2005, based on the emission-weighted sum for each sector. These state mortality cost factors reflect the mortality costs across the US of emissions from each state, regardless of where those deaths occur. For some states (e.g. California), most of the health impacts occur in the same state as the emissions. In other states, substantial health impacts may occur in downwind populous states. Growth factors were derived to account for changes in population exposure and baseline mortality rates in future years (Heo et al 2016a). Using these growth factors, the 2005 per-tonne mortality cost estimates are adjusted here for each GCAM-USA year (2010–2050 in 5 year increments). VSL also grows over time as a function of GDP per capita, reflecting increased willingness-to-pay (SI section 3). The 95% CI on total PMMC is quantified by applying RR adjustment factors (Heo et al 2016a) based on the 95% CI of RR (1.04–1.08) from Krewski et al (2009), reflecting the uncertainty in concentration-response function but not other potential sources of uncertainty (SI section 3).

Besides EASIUR, several other reduced-complexity models have been used to estimate regional PM_2.5 mortality coefficients, with relatively good agreement for national and sectoral estimates (Gilmore et al 2019). Here we use PM_2.5 mortality coefficients derived from EASIUR because it fits better with the sectoral-, spatial-, and temporal resolutions of GCAM-USA compared with other publicly available reduced-form approaches (SI section 3).

To quantify the relative importance of various factors to future PMMC changes, a decomposition approach is employed using the following two steps: First, analogous to the Kaya identity (Kaya 1990, Raupach et al 2007), the PMMC for a given state and year is expressed as the product of the following driving factors (table 1): population (${D}^{{P}{O}{P}}$), GDP per capita (${D}^{{\rm{G}}{\rm{D}}{\rm{P}}}$), sectoral energy intensity (${D}_{s}^{ENERGY}$), fuel share (${D}_{f,\,s}^{FUEL}$) and PMMC intensity (${D}_{f,\,s}^{{\rm{P}}{\rm{M}}}$) (equation (2)). Energy intensity (EJ/$) is the energy required to produce one unit of GDP, a measure of the energy efficiency of an economy. Fuel share (ratio) indicates the share of polluting fuels (coal, gas, biomass, and liquids) in each sector. PMMC intensity ($/tonne) represents the PMMC per unit energy use for a technology, as the product of emission intensity (tonne/EJ) and per-tonne mortality costs ($/tonne).

$$\begin{eqnarray}&&\begin{array}{l}M=\displaystyle \sum _{f}\displaystyle \sum _{s}{M}_{f,s}=\displaystyle \sum _{f}\displaystyle \sum _{s}P\,\times \displaystyle \frac{G}{P}\times \displaystyle \frac{{E}_{s}}{G}\times \displaystyle \frac{{E}_{f,s}}{{E}_{s}}\\ \,\,\times \,\displaystyle \frac{{M}_{f,s}}{{E}_{f,s}}=\displaystyle \sum _{f}\displaystyle \sum _{s}{D}^{POP}\times {D}^{{\rm{G}}{\rm{D}}{\rm{P}}}\times {D}_{s}^{ENERGY}\\ \,\,\times \,{D}_{f,s}^{FUEL}\times {D}_{f,s}^{{\rm{P}}{\rm{M}}}.\end{array}\end{eqnarray} \tag{ 2 }$$

Table 1.  Variables in LMDI decomposition.

Variable	Definition
f	Fuels that directly contribute to PM2.5 pollution, including coal, gas, refined liquids and biomass
s	Energy sectors, including electricity, industry, transportation, and building
M	Total PMMC for a given state and year
${M}_{f,s}$	PMMC attributed to fuel f in sector s for the state
P	State population (table S3)
G	State GDP (table S4)
${E}_{s}$	Total state energy consumption in sector s
${E}_{f,s}$	State energy consumption of fuel f in sector s
${D}^{POP}$	State population, same as P
${D}^{{\rm{G}}{\rm{D}}{\rm{P}}}$	State GDP per capita
${D}_{s}^{ENERGY}$	State energy intensity in sector s
${D}_{f,s}^{FUEL}$	Share of fuel f in sector s for the state
${D}_{f,s}^{{\rm{P}}{\rm{M}}}$	PM2.5 mortality intensity of fuel f in sector s for the state

Next, the logarithmic mean divisia index (LMDI) approach is adopted to separate the impacts of different factors on the overall changes in PMMC (equation (3)). LMDI has been widely employed for analysis of driving forces and to provide policy-relevant insights, including wind energy supply in China (Lu et al 2016) and emission trends of CO₂ in China (Guan et al 2018) and SO₂ in the US (Lu et al 2012). LMDI has been shown to have advantages over other decomposition methods because of its path independence, consistency in aggregation, and ability to handle zero values (Ang 2005, Ang and Wang 2015).

The changes in PMMC for a given region from year ${t}_{0}$ to year ${t}_{1}$ (${\rm{\Delta }}M$) can be expressed as the sum of contributions from individual factors:

$$\begin{eqnarray}&&{\rm{\Delta }}M=\displaystyle \sum _{k}{\rm{\Delta }}{M}_{k}=\displaystyle \sum _{k}\displaystyle \sum _{f}\displaystyle \sum _{s}L\left({w}_{f,s}^{{t}_{0}},{w}_{f,s}^{{t}_{1}}\,\right)\,\mathrm{ln}\left(\displaystyle \frac{{D}^{k,{t}_{0}}}{{D}^{k,{t}_{1}}}\right),\end{eqnarray} \tag{ 3 }$$

where k = POP, GDP, ENERGY, FUEL, PM and $L\left({w}_{f,s}^{{t}_{0}},{w}_{f,s}^{{t}_{1}}\right)=\left({M}_{f,s}^{{t}_{0}}-{M}_{f,s}^{{t}_{1}}\right)$/$\left({\rm{ln}}\left({M}_{f,s}^{{t}_{0}}\right).\,{\rm{ln}}\left({M}_{f,s}^{{t}_{1}}\right)\right)$ is the logarithmic mean of the mortality costs of fuel f in sector s at times ${t}_{0}$ and ${t}_{1}.$

Note that LMDI is fundamentally different from linear regression. For example, in LMDI, the change in the quantity being decomposed is always a linear combination of all contributors, with the residuals being allocated to contributions from individual factors (Lu et al 2012, Ang and Liu 2007). In contrast, linear regression typically produces a residual term that represents both model misspecification and confounding factors. Furthermore, LMDI estimates the total effect of each contributor on the PMMC changes, while regression coefficients represent the marginal effect of each contributor at their average level.

Between 2015 and 2050, emissions of primary PM_2.5, NO_x, and SO₂ are projected to decrease by 25%, 50%, and 17%, respectively (figure 1(a)). Sector-specific pollutant emissions projections are shown in figure S2. The corresponding mortality costs from emissions of primary PM_2.5, NO_x, and SO₂ decrease by 15%, 39% and 43%, respectively (figure 1(b)). The discrepancy between emission and mortality cost changes is a result of pollutant- and source category-specific mortality intensity trends (Heo et al 2016a), which are simultaneously affected by decreasing baseline mortality rates, increasing population exposure, and increasing VSL in the future. For all modeling years, primary PM_2.5 emissions are the greatest source of PMMC. The combined damages from NO_x and SO₂ are roughly equivalent to those of primary PM_2.5 in 2015, but by 2050 play a lesser role.

Zoom In Zoom Out Reset image size

Driven by existing regulations and projected demographic, economic, and technological changes, the national total PMMC (figure 2(a)) decreases 25% from $380 billion in 2015 to $284 billion in 2050. Overall, coal combustion sources maintain a relatively constant share of around 33% of PMMC for all modeling years, with major contributions transitioning from electric sector coal to industrial coal. The contribution of liquid fuels to PMMC drops from 35% in 2015 to 29% in 2050, dominated by the decreasing share attributed to transportation liquids.

Zoom In Zoom Out Reset image size

Among the expected drivers of PM_2.5 mortality trends, energy consumption increases by 12% in 2050 (figure 2(b)), but at a much slower rate than the growth in population (24%) and GDP per capita (52%) (figure 2(c)). The specific source sector-fuel combinations of industrial coal and building sector biomass are major contributors to PMMC (figure 2(a)) despite being minor contributors to overall energy consumption (figure 2(b)). Industrial sources often are not subjected to as stringent emission limits as is the electric sector. Thus, a quantity of coal combusted in the industrial sector typically produces more PM_2.5 and precursor emissions than if combusted in the electric sector. Building sector biomass largely consists of wood fireplaces and furnaces, which have high primary PM_2.5 emission intensities (figure 2(d)).

Most of the energy consumption in the building and industry sectors comes from gas and electricity. Over the modeling period, coal consumption for electricity production decreases from 12 EJ to 9 EJ (−25%), while the total energy consumption in the electricity sector increases from 30 EJ to 35 EJ (+17%). Together, these trends reflect that more electricity is being produced by clean (i.e. low or zero pollutant emitting) technologies in the future. Figure 2(b) also indicates that liquids remain the dominant fuel in the transportation sector for all modeling years, though liquids' share of total transportation energy consumption decreases from 99% in 2015 to 86% in 2050, replaced by electric and hydrogen vehicles.

The representation of air pollution regulations in GCAM-USA reduces PM_2.5 mortality intensity (mortality cost per unit energy input) by decreasing future emission factors (figure 2(d)). Due to regulations limiting coal-fired power plant emissions and Tier 3 motor vehicle and fuel emissions limits, PM_2.5 mortality intensities of electricity coal and transportation liquids decrease by 40% and 36%, respectively, contributing to their decreasing shares observed in figure 2(a). The PM_2.5 mortality intensity of building biomass decreases slightly before 2035, reflecting a gradual turnover from existing wood furnaces to more efficient, EPA-certified wood furnaces and stoves. Note that the temporal trend of PM_2.5 mortality intensity is also affected by the increasing VSL and population, as well as by changes in baseline mortality rates (Heo et al 2016a). For building sector biomass, the increasing trend of PM_2.5 mortality intensity after 2035 indicates that the effect of efficiency improvements is overcome by the increase in population exposure.

For the 25% national decrease in the central estimation of PMMC (figure 2(a)), LMDI decomposition reveals that the largest contributor is decreased energy intensity, followed by reduced PM_2.5 mortality intensities in electric sector coal and transportation liquids (figure 3). In the absence of other factors, these three factors would cause mortality costs to decrease by 68%. These beneficial contributions offset the increases in population and per capita GDP, which in isolation would increase PMMC by 54%.

Zoom In Zoom Out Reset image size

For electricity coal and transportation liquids, contributions from their decreased PMMC intensity is much greater than from their decreasing fuel shares. The trend of electricity coal is mainly due to CSAPR, which limits electric sector NO_x and SO₂ in the eastern US and the New Source Performance Standards (NSPS), which requires all new coal plants to apply controls on NO_x, SO₂, and PM_2.5 emissions. Within GCAM-USA, CSAPR is represented by state-specific, electric sector emission constraints, while NSPS is implemented with systematically lower emission factors applied in future modeling years. In the transportation sector, Tier 3 engine and fuel standards are reflected in the model as lower pollutant emissions per unit of fuel consumption over time. Independent of other factors, decreases in the PM_2.5 mortality intensities of electricity coal and transportation liquids each result in mortality cost decreases of approximately 12%. Changes in building sector biomass produce a net reduction in PMMC, resulting from a decrease in the share of building biomass, contributing −6%, and a small increase in its PM_2.5 mortality intensity.

Separate LMDI decompositions are further conducted to attribute changes in PMMC to changes in emissions of primary PM_2.5, NO_x, and SO₂ (figures S3–S5). As with the combined results for changes in emissions of all three pollutants (figure 3), decreases in species-specific mortality costs are predominantly due to decreases in energy intensity and are partially offset by increases in population and per capita GDP. However, the second most important drivers of decreasing mortality costs vary by species: building biomass fuel share (causing species-specific mortality costs to decrease by 11.8%, independent of other factors) for primary PM_2.5, PM_2.5 mortality intensity of transportation liquids (32.6% decrease) for NO_x, and PM_2.5 mortality intensity of electricity coal (45% decrease) for SO₂. These differences in main contributors further highlight the importance of tailoring air quality management strategies to address specific sectors and pollutants.

In addition to decomposing changes in national PMMC, characterizing changes in state-level PMMC and the underlying drivers may be even more important for state and regional energy planning and emission control strategy development. PMMC changes in 2050 vary significantly by state, ranging from −62% (ND) to +37% (FL) (figure 4).

Zoom In Zoom Out Reset image size

For most states, economic development, as measured by per-capita GDP growth, is the leading contributor to increased PMMC, while decreased energy intensity plays the dominant role in lowering PMMC (table 2). The second largest factor lowering PMMC for most states is the decreased PM_2.5 mortality intensity of either electricity coal or transportation liquids. Twelve of the 16 states whose top two contributors are related to electricity coal are CSAPR states.

Besides these general patterns, some states have specific challenges in reducing PMMC. For example, CT, ME, RI, and VT have high heating demand coupled with significant historical use of biomass. Consequently, increases in the share of biomass in the building sector in these states is the second largest contributor to their increased PMMC.

The 48 contiguous US states are categorized into two groups based on the shape of their projected PMMC trends (figure 5). The 'Steadily-decreasing' group includes 29 states, with monotonically decreasing PMMC, mostly located in the central US. Of these, Ohio and Pennsylvania have the highest PMMC in 2015. The 'Upturn' group includes the remaining 19 states, which have an increasing trend of PMMC either over the entire 2015–2050 period (such as Florida) or starting from an intermediate year (such as Arizona). From a policymaker's perspective, the Upturn states may indicate state-specific challenges to reducing PMMC in the future, which should be anticipated in the development of long-term air quality management strategies.

Zoom In Zoom Out Reset image size

In comparing the two groups, the Steadily-decreasing states have significantly (p < 0.001) greater decreases in state-level PMMC, smaller increases associated with population and per capita GDP growth, and greater decreases in electric sector coal share and PM_2.5 mortality intensity in 2050. This comparison highlights that population and economic growth can be dominant factors contributing to increased PMMC, especially for the Upturn states. Changes in electricity coal shares and PM_2.5 mortality intensity have significantly smaller effects among the Upturn states, because these states tend to have smaller shares of electricity coal in 2015. In addition, 8 of these 19 states are within the CSAPR region and their coal fleet emissions largely have been controlled by 2015, leaving less benefit available from turnover to cleaner coal capacity. Contributions from PMMC intensity of transportation liquids are similar among the two groups of states since Tier 3 tailpipe and fuel standards are national policies. Among states with little benefits from decreasing coal use, such as Florida, decreases in the PMMC intensity of transportation liquids play a major role in partially offsetting the effects of population and economic growth.