A decade of the U.S. energy mix transitioning away from coal: historical reconstruction of the reductions in the public health burden of energy

As the NEI is the most complete inventory of emissions in the U.S., it serves as the basis of our main analysis. The NEI is a bottom-up inventory of air pollutant emissions within the U.S. published by the U.S. EPA every 3 years [10]. The U.S. EPA constructs the NEI using emissions estimates provided by states, which are constructed using a mix of economic activity estimates, source data, energy consumption data, and monitoring data for larger sources (including CAMD data for electric generators ⩾ 25 MW) [10]. For this analysis, we used county-level data for fuel combusting sectors listed as non-mobile by the EIS, stationary sources related to the fuel supply chain, and other stationary industrial processes for the years 2008, 2011, 2014, and 2017. We then grouped each EIS sector into source categories with known fuel use (electricity, commercial buildings, residential buildings, and industrial boilers) as 'major fuel consuming stationary sources', and without fuel use or with unknown fuel type used (commercial cooking, waste disposal, supply chain, and other industry), as 'other stationary sources'. Major fuel consuming stationary sources were grouped by NEI fuel type (coal, gas, oil, wood, biomass, and other) (table S1). We combined wood and biomass into one category. We also use emissions data from CAMD, which provides SO₂ and NO_x emissions for some electricity generation, from 2008 to 2019 for electricity generation [15].

The RCMs (EASIUR, AP2, and InMAP) provide estimates of the total mortality impacts of PM_2.5 exposure per ton of each precursor pollutant—SO₂, NO_x , VOCs (except EASIUR), ammonia (NH₃), and primary PM_2.5—for each source county, from low and high stack heights [8, 10–12, 16]. We used the annual average values and assigned low, high, or the average stack heights to each source based on sector (table S1). We used a concentration response function (CRF) of a 1.4% increase in mortality risk per 1 µg m⁻³ change in the annual average PM_2.5 exposure [17]. This is a standard value from the RCMs, based on a CRF from a large cohort study in the epidemiological literature, which falls within the confidence intervals of a recent meta-analysis [3]. We used a value of statistical life (VSL) of $11.2 million (2017 USD) [18, 19]. We estimated the impact per ton of sources listed as 'portable facilities' in the NEI using the state's average for each RCM. To estimate the impact of sources on land currently controlled by Indigenous Americans, we assigned each source impact per ton values corresponding to the county or the average of all the counties in which land controlled by Indigenous Americans was located (table S2).

For each of the major fuel consuming stationary sources in the EIS—industrial boilers, commercial and residential buildings, and electricity generation—in each state, we developed state-level mortality impact factors (HIFs), providing mortality impacts per unit of energy consumption. To calculate the HIFs, we matched the mortality impacts for each state, year, source, and fuel type to energy consumption data in the corresponding state, year, source, and fuel type from SEDS and then calculate the HIFs using equation (2).

$$\begin{align} & {\text{Health}}{}\, {\text{Impact}}{{\text{s}}_{{\text{yea}}{{\text{r}}_i},{} {\text{stat}}{{\text{e}}_j},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}}\left( \$ \right) \nonumber \\[-4pt] & \quad = \sum\limits_{k = 1}^n {{\text{Health}}{}\, {\text{Impact}}{{\text{s}}_{{\text{yea}}{{\text{r}}_i},{} {\text{count}}{{\text{y}}_k},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}}\left( \$ \right)} \end{align} \tag{ 1 }$$

$$\begin{align} & {{\text{HI}}{{\text{F}}_{{\text{yea}}{{\text{r}}_i},{\text{ stat}}{{\text{e}}_j},{\text{ sourc}}{{\text{e}}_x},{\text{ fuel typ}}{{\text{e}}_y}{}}}\left( {\frac{\$ }{{{\text{quad}}}}} \right) } = {\frac{{{\text{Health Impacts}}{{ }_{{\text{yea}}{{\text{r}}_i},{\text{ stat}}{{\text{e}}_j},{\text{ sourc}}{{\text{e}}_x},{\text{ fuel typ}}{{\text{e}}_y}{ }}}\left( \$ \right)}}{{{\text{Primary Energy Consumptio}}{{\text{n}}_{{\text{yea}}{{\text{r}}_i},{\text{ stat}}{{\text{e}}_j},{\text{ sourc}}{{\text{e}}_x},{\text{ fuel typ}}{{\text{e}}_y}{ }}}{ }\left( {{\text{quad}}} \right)}}} \end{align} \tag{ 2 }$$

where, ${\text{Health}}{} {\text{Impact}}{{\text{s}}_{{\text{yea}}{{\text{r}}_i},{} {\text{stat}}{{\text{e}}_j},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}}$ is the health impacts of emissions from ${\text{fuel}}{} {\text{typ}}{{\text{e}}_y}$ combusted by ${\text{sourc}}{{\text{e}}_x}$ (major fuel consuming stationary sources of electricity, commercial buildings, residential buildings, and industrial boilers) in ${\text{stat}}{{\text{e}}_j}$, in ${\text{yea}}{{\text{r}}_i}$; ${\text{Health}}{} {\text{Impact}}{{\text{s}}_{{\text{yea}}{{\text{r}}_i},{} {\text{count}}{{\text{y}}_k},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}}$ is the health impacts of emissions from ${\text{fuel}}{} {\text{typ}}{{\text{e}}_y}$ combusted by ${\text{sourc}}{{\text{e}}_x}$ (major fuel consuming stationary sources) in ${\text{count}}{{\text{y}}_k}$, in ${\text{yea}}{{\text{r}}_i}$; and there are k = 1, ..., n counties in ${\text{stat}}{{\text{e}}_j}$; ${\text{Primary}}{} {\text{Energy}}{} {\text{Consumptio}}{{\text{n}}_{{\text{yea}}{{\text{r}}_i},{} {\text{stat}}{{\text{e}}_j},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}}$ is the energy consumption, with unit of quad (1 quad equals 1 quadrillion, 10¹⁵, Btu), in ${\text{fuel}}{} {\text{typ}}{{\text{e}}_y}$ by ${\text{sourc}}{{\text{e}}_x}$ in ${\text{stat}}{{\text{e}}_j}$ in ${\text{yea}}{{\text{r}}_i}$.

We then project health impacts from energy use reported by SEDS, using HIFs calculated from the most recent prior year available (i.e. impacts estimated using the 2014 NEI were divided by 2014 energy use from SEDS to produce state level estimates of emissions factor (tons per unit of energy) and HIFs for 2014; the 2014 emissions factors and HIFs were then used to estimate impacts for 2015 and 2016 using SEDS data from the corresponding years).

$$\begin{align} & {\text{HI}}{{\text{F}}_{{\text{yea}}{{\text{r}}_i},{} {\text{stat}}{{\text{e}}_j},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}{} }} \nonumber\\ & \quad \approx {\text{HI}}{{\text{F}}_{{\text{yea}}{{\text{r}}_{i + m}},{} {\text{stat}}{{\text{e}}_j},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}} \end{align} \tag{ 3 }$$

$$\begin{align} & {\text{Health Impact}}{{\text{s}}_{{\text{yea}}{{\text{r}}_{i + m}},\;{\text{stat}}{{\text{e}}_j},\;{\text{sourc}}{{\text{e}}_x},\;{\text{fuel typ}}{{\text{e}}_y}\,}}\left( \$ \right) = {\text{HI}}{{\text{F}}_{{\text{yea}}{{\text{r}}_i},\;{\text{stat}}{{\text{e}}_j},\;{\text{sourc}}{{\text{e}}_x},\;{\text{fuel typ}}{{\text{e}}_y}\,\,}}\left( {\frac{\$ }{{{\text{quad}}}}} \right) \nonumber \\ & \quad \times {\text{Primary Energy Consumptio}}{{\text{n}}_{{\text{yea}}{{\text{r}}_{i + m}},\;{\text{stat}}{{\text{e}}_j},\;{\text{sourc}}{{\text{e}}_x},\;{\text{fuel typ}}{{\text{e}}_y}}}\,\left( {{\text{quad}}} \right) \end{align} \tag{ 4 }$$

where ${\text{Health}}{}\, {\text{Impact}}{{\text{s}}_{{\text{yea}}{{\text{r}}_{i + m}},{} {\text{stat}}{{\text{e}}_j},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}}$ is the health impacts projected in ${\text{yea}}{{\text{r}}_{i + m}}$;  Primary ${\text{Energy Consumption}}{{\text{}}_{{\text{yea}}{{\text{r}}_{i + m}},{} {\text{stat}}{{\text{e}}_j},{} {\text{sourc}}{{\text{e}}_x},{} {\text{fuel}}{} {\text{typ}}{{\text{e}}_y}}}$ is the energy consumption in ${\text{yea}}{{\text{r}}_{i + m}}$; and m (1, 2, and 3) represents the years in which the health impacts are projected based on the most recent HIF calculated using the most recent available historical emissions data from ${\text{yea}}{{\text{r}}_i}$ and the corresponding energy use data in ${\text{yea}}{{\text{r}}_i}$.

We then compared the results of this projection method against the next available estimates derived from the NEI (i.e. we estimated 2017 impacts using 2017 SEDS data and a 2014 state-level HIFs, and then compared it to estimates using the 2017 NEI emissions data), using percent difference between the NEI-based historical estimate and the corresponding projected impact.

$$\begin{align} & {\text{Health}}{} {\text{Impact}}{} {\text{Projection}}{} {\text{Error}}{} \left( \% \right) \nonumber\\ & \quad = \frac{{\begin{array}{*{20}{c}} {({\text{Heath}}{} {\text{Impac}}{{\text{t}}_{{\text{yea}}{{\text{r}}_{i + m,{\mkern 1mu} {\text{projected}}}}}}} \\ { - {\text{Heath}}{} {\text{Impac}}{{\text{t}}_{{\text{yea}}{{\text{r}}_{i + m,{\mkern 1mu} {\text{historical}}}}}})} \end{array}}}{{{\text{Heath}}{} {\text{Impac}}{{\text{t}}_{{\text{yea}}{{\text{r}}_{i + m,{\mkern 1mu} {\text{historical}}}}}}}} \times 100\end{align} \tag{ 5 }$$

where ${\text{Health}}{} {\text{Impact}}{} {\text{Projection}}{} {\text{Error}}$ is the percent difference between ${\text{Heath}}{} {\text{Impac}}{{\text{t}}_{{\text{yea}}{{\text{r}}_{i + m,{\mkern 1mu} {\text{projected}}}}}}$ (the projected health impacts in ${\text{yea}}{{\text{r}}_{i + m}}$ using HIF calculated from emission and energy data in ${} {\text{yea}}{{\text{r}}_i}$) and ${\text{Heath}}{} {\text{Impac}}{{\text{t}}_{{\text{yea}}{{\text{r}}_{i + m,{\mkern 1mu} {\text{historical}}}}}}$ (the historical health impacts in ${\text{yea}}{{\text{r}}_{i + m}}$ using historical emission data in ${\text{yea}}{{\text{r}}_{i + m}}$).

Positive and negative Health Impact Projection Error values represent overestimation and underestimation of the estimated health impacts using the state-level projection method, compared to estimates derived from the county-level historical data from the NEI. For this comparison, we are treating the health impacts estimated from historical NEI data as the more accurate model for the basis of comparison.

The mortality impacts of PM_2.5 from major fuel combusting stationary sources have changed substantially in the last decade. The trend was consistent across RCMs, but with some disagreements in magnitude. All ranges provide the range of estimates from the three RCMs with our chosen CRF and VSL. Impacts of electricity had the greatest reduction, from 59 000–66 000 deaths ($660—$740 billion) in 2008 to 10 000–12 000 deaths ($110—$140 billion) in 2017 (figure 1, table 1), largely driven by reduced emissions from coal (figures 2 and S1(a)). Sources categorized as 'Other Industry' had the second largest reduction, from 41 000–49 000 attributable deaths ($460—$550 billion) in 2008 to 27 000–32 000 attributable deaths ($300—$350 billion) in 2017 (figure 3, table 1), largely driven by reductions in impacts from SO₂ and PM_2.5 emissions (figures 4 and S2(b)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Industrial boilers and commercial buildings had substantial decreases in coal and oil impacts, mostly from SO₂, but these were essentially replaced by increases in biomass combustion impacts, mainly PM_2.5 emissions (figures 2 and S1(a)). Impacts from commercial building gas use peaked in 2011, and then decreased from 2011 to 2017; industrial boiler gas use impacts consistently decreased from 2008 to 2017 (figures 2 and S1(a)). Biomass and wood combustion was the largest contributor to health impacts from residential buildings, largely driven by primary PM_2.5 emissions (figures 1 and 2), but impact from this source decreased from 2008 to 2017. Gas use was the second-largest contributor to health impacts from residential buildings, largely driven by NH₃ emissions (figures 1 and 2). The two main contributing emitted pollutants were NO_x , which had peak impact in 2011, and PM_2.5, which had peak impact in 2014 (figures 1 and 2).

Other stationary sources generally had decreasing impact over time, roughly paralleling decreasing emissions (figures 3 and S1(b)). All industrial processes except for oil and gas production had decreasing impact over time, largely driven by reductions in SO₂ and PM_2.5 emissions (figures 3, 4 and S1(b)). The highest of these sectors—industrial processes not elsewhere classified—had a health burden of 16 000–20 000 excess mortality cases ($180—$230 billion) in 2008, which decreased to 12 000–14 000 mortality cases ($130—$150 billion) in 2017 (figures 5, 6 and table 3). Waste disposal impacts varied substantially year to year but stayed between 6900 and 11 000 mortality cases, mostly from changes in emissions of PM_2.5 and NH₃ (figures 3, 4 and table 3). Commercial cooking impacts consistently increased from 2008 to 2017, with 4800–8200 mortality cases ($53—$92 billion) in 2008, increasing to 7100–13 000 mortality cases ($80—$150 billion) in 2017 (figures 5, 6, S1(b) and table 3).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Agreement between RCMs tended to be the highest for major fuel combusting stationary sources, especially coal, and was higher in 2008 compared to 2017. From 2008 to 2017, disagreement between RCMs increased for all sources except Other Industry (table 1). InMAP tended to give the highest estimates—7 of 8 in 2008, and 5 of 8 in 2017; AP2 tended to give the lowest, providing the lowest values for 5 of the 8 sectors in 2008, and 7 of 8 in 2017 (table 1). The agreement between RCMs decreased from 2008 to 2017 for 14 of the 17 sector and fuel type combinations (table 2). For sectors that had coal as a fuel type, the percent difference between RCMs was between 13% and 19% (table 2). The percent difference between RCMs for gas, oil, biomass, and all other fuel types ranged between 29% and 46%, 23% and 35%, 45% and 55%, and 4.8% and 41%, respectively (table 2). There was lower agreement between the RCMs for the non-combustion sources, and less of a consistent pattern between models (table 3). AP2 or EASIUR tended to provide the lowest value, while the highest values were more evenly shared between RCMs. The percent differences also had a much wider range for these sources, especially for sources dominated by VOC emissions, since EASIUR does not provide estimates for VOCs (table 3) [11, 12].

In 2017, the highest contributing sectors were industry, industrial boilers, and residential heating, with electricity and commercial buildings both now dropping to 4th highest (table 1). In 2017, the impacts from industrial boilers were largely from PM_2.5 emissions from biomass combustion, followed by both SO₂ from coal and NO_x from gas (figure 5). The highest contributor to the impact from residential heating in 2017 was emissions of mainly PM_2.5 emissions from biomass and wood, followed by a mix of NO_x , NH₃, and PM_2.5 emitted from gas (figure 5). Electricity impacts were still dominated by coal emissions, largely from SO₂. Commercial building impacts were driven mainly by emissions of NO_x , PM_2.5, and some NH₃ from gas, followed by PM_2.5 from biomass (figure 5). For non-combustion stationary sources, the three sectors with the highest impacts were the other industrial processes (mainly from PM_2.5 emissions), commercial cooking (entirely PM_2.5 emissions), and waste disposal (mainly PM_2.5 emissions) (table 3, figure 6).

The highest impact fuel type varied substantially by state in 2017. All three RCMs indicate that coal was the highest impact fuel in 12 states, biomass and wood in 24, and gas in 3 (figure 7), with model disagreement in 11 states. When broken down by source, in most states, coal is the highest contributor to electricity generation impacts, gas and biomass and wood are the highest contributors to impacts in both commercial buildings and industrial boilers, and biomass and wood are the highest contributor for residential buildings (figure 8).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The highest impact non-combustion sector varied substantially between states—industrial processes not elsewhere classified had the highest impact in 11 states and waste disposal and commercial cooking were each highest impact in 5 states. The RCMs disagreed in 22 states (figure 9).

Zoom In Zoom Out Reset image size

Across sources, the total mortality impacts of coal have decreased substantially from 2008 to 2017. While trends in impacts of gas combustion varied across sources, total health impacts from gas were slightly lower than that of coal in 2017, with the two essentially converging in 2018 (table 2, figure 10). In 2018, health impacts from coal are projected to be $150-$170 billion, while gas impacts are projected to be $130-$170 billion (figure 10), with InMAP projecting that health impacts from gas exceed that of coal. Between 2008 and 2017, impacts from coal dropped substantially in commercial buildings, industrial boilers, and electricity, and were negligible in residential buildings. Gas impacts remained roughly constant in buildings and electricity, but decreased in industrial boilers (figure 11). The state-level projections, when compared with impacts based on historical data from the NEI, perform better at a national level compared to state level. On an aggregate national level, projection errors range between ∼76% underestimation and ∼88% overestimation (figure S2), while projection errors for state-level health impacts range between ∼100% underestimation and ∼390% overestimation, for coal, gas, and biomass and wood (figures S3 and S4).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Health impacts modeled using the CAMD dataset are lower than those using the NEI, in part due to NEI having more comprehensive coverage of electricity generation emissions (figure 12) [10, 15]. Nevertheless, CAMD results do indicate a further decrease in 2018 and 2019 in health impacts from coal-fired power plants emissions, supporting the results from our state-level projections. All three RCMs indicate that gas had higher health impacts than coal in 8 states in 2008, and that number increased to 20 in 2017, largely on the eastern and western coasts of the U.S. (figure 13). In 2017, emissions from gas combustion had higher impact than those from coal in all 48 states, the District of Columbia, and land currently controlled by Indigenous Americans for residential buildings, 45 states for commercial buildings, 29 states for industrial boilers, and 10 states for electricity (figure 14). Across fuel types and major fuel combusting sources, the RCMs results were within a factor of 2 to an order of magnitude of each other, and this was consistent across county to state to nationwide levels of aggregation (figures 15, 16, and S7).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size