Modeling energy system and air quality impacts of electrification
The U.S. Regional Economy, GHG, and Energy (US-REGEN) framework combines a state-of-the-art electric sector capacity planning and dispatch model with a uniquely capable end-use model40,41. Distinguishing features of the model include: (1) Detailed disaggregation of end-use sectors, activities, end-uses, and technologies and explicit tracking of structural classes including building type and size, building and equipment vintage, household attributes, and annual temperature profile; (2) endogenous end-use technology adoption; and (3) synchronized equilibrium of hourly load profiles and prices between electricity supply and energy use. These features enable US-REGEN to systematically represent many important dimensions of end-use technology tradeoffs, such as the heterogeneity of applications and customers, which better captures economic, behavioral, and policy factors influencing electrification and consequently emissions. Moreover, the integrated representation of electricity supply and demand provides a dynamic and scenario-consistent treatment of the marginal emissions from increased electric generation to support electrification, taking into account structural changes to the generation mix over time rather than relying on a historical snapshot, as many marginal emissions studies of electrification do20. Note that the energy system model scope for this analysis does not include endogenous representations of fossil fuel prices, biofuel production, non-electric fuel movement, or general equilibrium effects. US-REGEN is documented in detail in EPRI (2020)41, so only summaries of key features are provided in the Methods and Supplemental Information.
We consider three scenarios that differ in the extent of electrification and drivers of decarbonization:
Limited electrification: This scenario assumes limited electric vehicle (EV) adoption for light-duty vehicles, no growth in building electrification, and no concerted federal climate policy. These assumptions lead electricity demand to be approximately flat over time. This hypothetical benchmark was constructed to display conservative levels of electrification that, when compared to results from other scenarios, help to better quantify the impact of electrification on outcomes of interest.
High electrification without carbon price: This scenario includes more optimistic assumptions about advanced end-use technologies but does not include national CO2 policies. Higher electrification in this scenario is driven by: (1) allowing EVs to deploy endogenously (given continuing trends of falling battery costs); (2) accelerating the performance assumptions for heat pumps; and 3) faster depreciation of the existing equipment stock. This scenario more closely approximates expected market trends than does the Limited Electrification case, which allows us to evaluate the effects of such electrification.
High electrification with carbon price: This scenario has the same technology assumptions as the second scenario and introduces a carbon price starting in 2025 at $50/tCO2 (in 2020 USD) and growing at 7% per year ($271/tCO2 in 2050), which is intended as a proxy for a suite of CO2 policies for the electric and end-use sectors.
Detailed scenario assumptions are provided in Supplementary Note 6.
The US-REGEN model estimates CO2 and criteria air emissions from anthropogenic sources for the entire Continental U.S. (CONUS) for each scenario over the period 2015–2050 in five-year increments. We model air quality for four scenarios: (1) 2035 Limited Electrification; (2) 2035 High Electrification without Carbon Price; (3) 2035 High Electrification with Carbon Price; and (4) 2050 High Electrification with Carbon Price. Modeling these four scenarios provides a wide range of potential futures and probes the relative roles of electrification technology and carbon pricing to air quality. The criteria emission estimates are made for ozone and PM precursors, including NOx, SOx, volatile organic compounds (VOC), carbon monoxide (CO), ammonia (NH3), and primary PM. These emissions are processed to develop more temporally, spatially, and chemically refined inputs for air quality modeling. US-REGEN emissions by sector, source, fuel type, and region are cross-referenced to Source Classification Codes (SCC), county, and North American Industry Classification System (NAICS) code (see Supplementary Note 2 for more information).
We use the Comprehensive Air Quality Model with Extensions (CAMx), version 7.0, for air quality modeling with a 12-km grid covering the entire lower 48 states and nested within a larger 36-km grid (Supplementary Fig. S4) for every hour of the calendar year 2016. The meteorology for 2016 is used for all scenarios so that the projected air quality changes can be attributed solely to emissions changes. Our simulations build from a CAMx database developed by EPA and used for national air quality policy42.
Energy system and emissions results
Results in US-REGEN show how electrification can significantly raise electricity demand (Fig. 1) and electricity’s share of final energy (Supplementary Fig. 17). Electricity currently represents about 20% of final energy (similar levels to the Limited Electrification scenario in future years), which grows in the High Electrification scenario to 31% in 2035 and 34% in 2050 (and to 34 and 51% when CO2 policy is added), driven by technological change and further bolstered by CO2 policy. Electricity demand is roughly flat over time in the Limited Electrification scenario, as growing service demand is offset by efficiency increases. The High Electrification scenario entails 23% (39%) growth by 2035 (2050), which increases with CO2 policy to 24% (52%) by 2035 (2050). Electrification of transport (both light- and heavy-duty vehicles) and industry (e.g., process heat) represent the largest contributors to load growth (Fig. 1, top). Note that the levels of electric vehicle deployment in the High Electrification scenario more closely resemble anticipated electrification based on projected market trends (Supplementary Fig. 13). Electrification is also occurring in buildings, but efficiency improvements lead to roughly offsetting effects in terms of total electricity demand. These levels of electrification are comparable to other deep decarbonization studies in the literature (Supplementary Fig. 14).
The carbon intensity of electricity generation declines over time in all scenarios (Fig. 1, middle). In the absence of national CO2 policies, natural gas and renewables grow, while existing coal and nuclear are gradually retired. Unlike other studies, higher electrification does not lead to increases in dispatch from existing coal in these scenarios even without a decarbonization policy, as the endogenous investment and retirements lead to declining coal generation, which contrasts with dispatch-only short-run marginal emissions studies20. Load growth in the High Electrification scenario is largely met with expanded gas and wind generation, which has a significantly lower emissions profile, both in terms of CO2 and criteria pollutants, than the current generation mix. When carbon pricing is added, coal is phased out decades earlier; most natural gas is equipped with carbon capture and storage (CCS) or is co-combusted with hydrogen; nuclear remains in the mix; and solar and wind see much larger increases. Increasing the share of electricity in final energy consumption can facilitate reductions in CO2 and air pollutant emissions economy-wide, even with substantial electric load growth, particularly when combined with a shift away from combustion-based electricity generation (Supplementary Fig. 18).
CONUS-wide emissions are shown in Supplementary Figs. 3 and 4 and Table 1. Economy-wide CO2 emissions decline across all scenarios (Fig. 1, bottom), though the rate and extent vary by scenario. CO2 reductions relative to 2005 levels are 33% in the Limited Electrification scenario, 44% in the High Electrification, and 78% in the High Electrification with Carbon Price scenario. In April 2021, the U.S. updated its pledge as part of the Paris Agreement to reduce emissions by 50–52% by 2030 from 2005 levels43. As shown in Supplementary Fig. 19, the High Electrification with Carbon Price scenario is consistent with this target, while the other scenarios fall short, entailing 19–28% CO2 declines by 2030. NOx emission reductions are 28% from 2016 to the 2035 Limited Electrification scenario due to on-road and off-road fleet turnover and cleaner fleet mix in the future year. NOx emission reductions are accelerated in the 2035 High Electrification scenario for on-road and off-road sectors, even with growing population and economic activity, resulting in net NOx reductions of 58 and 46% with and without carbon pricing, respectively. Emission decreases from the electric sector result mainly from declining emissions from coal generation (Fig. 1, bottom). Under the High Electrification with Carbon Price scenario, electric generating unit (EGU) emissions drop significantly from the 2016 levels for SO2 (99%) and NOx (82–87%). VOC emission reductions from on-road vehicles and off-road sources are reduced more than 80% in the High Electrification scenarios, but they are offset by emission increases from oil and gas activity and other nonpoint sources, such as solvent utilization and commercial and residential fuel combustion. Total primary PM emissions, which are dominated by fugitive dust emissions from roads, agriculture, and construction, change only by 3–6% from 2016 (Supplementary Table 1). Dust emissions (except for paved road dust) are held constant at 2016 levels with an assumption that future controls will offset growth (consistent with EPA procedures in the 2016 emission inventory). Activity growth in the agriculture sector (livestock waste and fertilizer application) results in NH3 emission increases of about 20% in 2035 and 30% in 2050. Overall, categories of emissions that are dominated by fuel combustion exhibit larger declines with electrification and decarbonization (e.g., CO2, SO2, NOx) than emissions with substantial non-combustion sources (e.g., PM, NH3).
Air quality results
Air quality model results from CAMx are shown in Fig. 2 for the ozone design value (DV) metric under the NAAQS, i.e., the fourth highest MDA8 ozone concentrations, the form of the standard that with specified exposure levels is deemed protective of primary (human health) and secondary (welfare) effects, currently set at 70 ppb for both endpoints. DV changes in future years were calculated according to U.S. EPA guidance44 by using model results to adjust the historical DV for the baseline scenario. Deep NOx emission reductions (ranging from 28 to 67% from 2016) lead to striking ozone DV reductions across the CONUS, though the magnitude of response varies regionally. In particular, the geographic distribution of benefits is highest for states in the Northeast, Southeast, and Ohio River Valley. Ozone benefits due to vehicle fleet turnover in the 2035 Limited Electrification scenario are widespread leading to attainment of the 70 ppb NAAQS in the eastern states but not all of the western states (e.g., California). The western states experience higher background ozone originating from global natural (e.g., wildfires, biogenic, lightning NOx, and stratosphere-troposphere exchange) and international precursor sources45 making it more difficult for these areas to meet the NAAQS. Our discussion below focuses on regions outside California, as prior studies have addressed California with state-specific assumptions37,46. High Electrification without carbon pricing can further reduce ozone by 3–13 ppb in 2035. The impact of carbon pricing in 2035 is most evident in the Midwest and eastern Texas, which show additional ozone reductions of 6–10 ppb compared to 2–4 ppb in other areas. This is due to substantial decreases of NOx emissions from EGUs (75% nationally) and reduced oil and gas activities (19% nationally). Ozone in urban areas declines faster than in rural areas because on-road vehicles are key in driving the overall NOx emission reductions, and these sources are more concentrated in urban areas. However, total NOx emissions remain mostly higher in urban areas and peak ozone DV locations remain here in the future.
These scenarios show that electrification leads to air pollution benefits, though these benefits are amplified by carbon pricing policy. The air quality improvements shown in Fig. 2 are substantial even when power sector only moderately decarbonizes. This finding suggests that near-term substitution of fossil fuels for electricity could yield immediate benefits for human health, though the impact varies regionally and by end-use application. Benefits from improved air quality occur rapidly following emissions reductions with a response that is spatially close to where mitigation happens, which differs from benefits from decreases in climate damages that may take decades to be felt and may occur in geographically distant locations.
Historical ozone DVs show a declining trend because emissions of ozone precursors continue to decline from 2002 levels, but progress has been slowing in recent years or even reversed in some locations. An example is presented in Fig. 3 for the Aldine monitor in Houston (Texas). The ozone DV at Aldine declined from 100 ppb in 2003 to a low of 72 ppb in 2014 but plateaued around 80 ppb since then, i.e., a lack of progress towards attainment of the 70 ppb NAAQS. Other monitors in Texas and other areas show similar long-term ozone declines that became flat recently (Supplementary Fig. 8). On-the-books strategies lead to the projected 2035 DV of 74 ppb at Aldine and marginally below 70 ppb in West Alton (Missouri), Alsip (Illinois), Dallas (Texas), and San Antonio (Texas), shown in SI. The High Electrification scenario lowers 2035 DV to 67 ppb (7 ppb lower than the Limited scenario). Adoption of carbon pricing lowers ozone by another 4 ppb. The ozone benefit from carbon pricing in 2035 is about 10 ppb in northeast Texas (near Cotton Valley Sand and Haynesville Shale oil and gas activities) compared to 2–4 ppb in other areas. Increased levels of electrification can double or even triple ozone improvement from 2016 as seen in the 2035 and 2050 with carbon pricing scenarios. In 2050, many monitors such as Fulton (Georgia) and Capitol (Louisiana) approach levels that are considered background45.
Annual PM2.5 DV, defined as the annual mean averaged over three years, declined gradually from 2002 but increased more recently at some monitors (Supplementary Figs. 9–12). Current annual PM2.5 DVs (2018–2020) at monitors in the eastern U.S. meet the 2012 NAAQS of 12 μg m−3 but many are above 10 μg m−3. Harvard Yards, which is a controlling monitor in Cleveland (Ohio), had a 2017–2019 DV of 11 μg m−3. The 2035 Limited Electrification scenario reduces PM2.5 DV from 2016 by 0.3 μg m−3 at this monitor and only up to 0.5 μg m−3 elsewhere (Supplementary Figs. 9–11). PM2.5 DV increases from 2016 at some monitors such as at Clinton in Houston (Texas) and Granite City (Illinois). PM2.5 DV further decreases in the 2035 High Electrification without Carbon Price by 1.0 μg m−3 at Harvard Yards, however, most areas see PM2.5 DV decreases of <0.5 μg m−3. Carbon pricing further lowers PM2.5 DV by 0.7 μg m−3 at this monitor in line with 0.5–1.0 μg m−3 elsewhere. PM2.5 DVs in 2050 are similar to 2035 with a carbon price due to increases of primary PM2.5 emission from industrial sources (such as metal production and mineral products) offsetting decreases of PM2.5 precursor emissions (NOx, SO2).
PM is comprised of many chemical components including both organic and inorganic particles. Decreases in PM2.5 DV from 2016 are driven by reductions of both primary (elemental carbon) and secondary PM2.5 species (sulfate and ammonium), which decrease in the 2035 Limited Electrification and even more in high electrification scenarios (Supplementary Figs. 9–11). PM2.5 nitrate decreases in all high electrification scenarios but increases in the Limited Electrification scenario in wintertime along the Ohio River Valley, where many EGUs are situated and in Texas. Other modeling studies also have found that SO2 emission reductions can lead to PM2.5 nitrate increases in the eastern U.S. (e.g.,47) although we note that comparisons of decadal trends in modeled and observed PM2.5 suggest that models may overstate this effect, possibly because widely used algorithms for modeling the sulfate-nitrate-ammonium aerosol system are biased. If our modeling tends to overstate PM2.5 nitrate increases caused by SO2 reduction, then our estimated benefits of electrification will tend to be understated.
Organic aerosol decreases only marginally, and crustal material increases due to increase in the primary PM2.5 emissions. While the combustion-related PM2.5 emissions from electrified sources decrease, growing population and economic activity are expected to maintain or even increase emissions from multiple source categories such as road dust (tire, brake pad, and roadway treatment), commercial wood/biomass boilers, and industrial facilities (metal production and mineral products). Reductions of emissions from several activities in addition to combustion-related emissions may be needed to further reduce PM2.5. Although PM-related impacts are relatively small on a percentage basis, even small changes in PM2.5 concentrations can have a large response in health benefits19.
Insights gained from detailed modeling
This analysis finds that electrification decreases CO2 and other air pollutant emissions even in the absence of a national climate policy, and these declines are amplified by decarbonization policies. These findings contrast with earlier studies suggesting more limited emissions effects of electrification, especially those that use short-run marginal emissions estimates10,20, which characterize marginal emissions from fixed electricity systems and do not account for structural changes over time as many energy systems models do. To illustrate how detailed energy systems modeling facilitates these findings, this section compares our results with simplified approaches using emissions factors that are common in the literature and in actual decision-making48:
REGEN observed: Net emissions effects are based on outputs from a detailed structural model (REGEN) of electric sector capacity planning and dispatch as well as end-use adoption. These correspond to emissions changes in earlier sections. Many systems modeling methods endogenize CO2 emissions from the power sector (and all other sectors) and implicitly use this approach.
REGEN average emissions: Net effects based on annual average emissions rate outputs from REGEN. These dynamic emissions rates vary over time with changing resource mixes and by model region. Average emissions rates are often used for their simplicity49, though they omit changes that act on a system’s margin, where the generation response to new or existing loads may differ from the current average mix48.
Constant average emissions: Net effects use average emissions rates, which are assumed to remain constant over time. Annual 2019 values come from ref. 20 at the interconnect level.
Constant marginal emissions rates (annual): Net effects use short-run marginal emissions rates from ref. 20, which are assumed to remain constant over time. Annual 2019 values are used at an interconnect level. Short-run marginal emissions rates estimate how new loads would be served from the current grid but do not account for how new loads could induce investments or retirements of assets48.
Constant marginal emissions rates (hourly): Net effects use 2019 marginal emissions rates from ref. 20, which are reported at hourly levels.
Figure 4 shows CO2 emissions reductions from higher electrification across these different methods. Marginal and average emissions approaches systematically underestimate reductions from electrification: These emissions factors only capture 52 to 91% of anticipated CO2 reductions under a reference scenario and 32 to 74% of anticipated CO2 reductions under a carbon pricing scenario. Marginal emissions estimates implicitly assume that coal and gas generation are on the margin for large fraction of hours48 and do not capture changes over time. Average emissions metrics can better approximate anticipated emissions changes, especially dynamic ones based on model outputs that account for changes in the grid mix over time.
The emissions bias from simplified emissions factor methods is larger under scenarios where structural change is expected in the electric sector, especially with stringent CO2 policies. Structural modeling methods, such as the approach in this paper, are particularly important in settings where system changes are expected over the life of the asset. Constant average emissions perform better than marginal emissions but worse than observed reductions for a carbon tax, since the electricity generation fleet decarbonizes over time. Supplementary Fig. 20 illustrates the rapid decline in emission intensity of electricity generation over time in the carbon price scenario.
Overall, these comparisons indicate that emissions declines from electrification are larger than simplified marginal and average emissions methods suggest. Emissions decline with today’s grid mix but are lower with future market trends (e.g., coal retirements, renewables) and existing state policies. End-use electrification can be environmentally beneficial today and increase over the lifetime of the device.
Visit our sponsors
Wise (formerly TransferWise) is the cheaper, easier way to send money abroad. It helps people move money quickly and easily between bank accounts in different countries. Convert 60+ currencies with ridiculously low fees - on average 7x cheaper than a bank. No hidden fees, no markup on the exchange rate, ever.