Carbon Scoring: Initial Baseline

This analysis establishes an initial baseline for RFF’s carbon scoring work, comparing our current central projection, including the Inflation Reduction Act (IRA) and more recent policy developments, against a baseline scenario from 2022. We additionally evaluate sensitivities related to oil and gas prices, technology availability, and regulatory actions. Our results are summarized in the Carbon Scorecard below.

Explore these results in detail below using our interactive results browser. For more information about our approach and models, see the methodology and technical appendix sections. The supporting data are available for download below each figure, and the full source code for RFF’s Haiku model is available here.

Incentives put in place by the IRA are projected to produce substantial CO₂ emissions reductions and put the US economy on an initial pathway to net zero by 2050. These reductions come with health benefits and reduced premature death from associated improvements in air quality, reduced retail electricity prices, and overall are net beneficial to society.

IRA incentives for clean electricity are in place throughout the projection period. However, the pace at which emissions are reduced in the power sector slows over time under all scenarios explored. Reductions from the power sector are not matched by similar emissions decreases in other sectors, leading to an increasing deviation from a net zero pathway over time for the economy as a whole. This divergence suggests a need for additional policy measures to support decarbonization across all sectors of the economy to maintain an emissions trajectory consistent with full decarbonization by 2050.

The results provided in this carbon score are generated using multiple models operated in tandem by RFF. Effects on the power sector and light duty vehicles are estimated using Haiku and the Vehicle Market Model (VMM) respectively. Effects on human health from changes in emissions are calculated using Estimating Air Pollution Social Impact Using Regression (EASIUR).

Outputs are drawn from other models to cover sectors for which RFF does not currently operate sectoral models. Output from the National Energy Modeling System (NEMS) is used to provide calibration inputs as well as emissions projections for the industrial, medium- and heavy-duty transportation sector, and residential and commercial buildings. The technical appendix below provides additional detailed information about these models and how they are operated together. At present, feedbacks between models are not accounted for; each model has been run once for a given scenario. Our estimates are that the effects of such feedbacks on overall results would be minor.

This score provides comparisons among the following five scenarios:

2022 Baseline Scenario

The 2022 reference baseline incorporates federal policies that were in place by the end of 2022, with the exception of the Inflation Reduction Act. This includes:

Infrastructure Investment and Jobs Act (IIJA): The IIJA civilian nuclear program is modeled as an extension on the life of the Diablo Canyon plant in California. Incentives for electric vehicle charging infrastructure are also represented.

EPA Fuel Economy Regulations 2024-2026: These regulations target a fleet-wide average of 161 g CO₂/mile, estimated to result in average fuel economy labels (reflecting real-world driving conditions) of 40 mpg for model year 2026. The model maintains the 2026 targets for subsequent years. (Note that on March 20, 2024, EPA finalized new requirements for model years 2027 and after. The new requirements will be the subject of a future score and at that point will become part of the new baseline.)

State and regional policies: The Regional Greenhouse Gas Initiative (RGGI) cap-and-trade program for the power sector is represented with a cap-and-trade constraint that accounts for allowance banking and different price steps in the allowance supply curve for the emissions containment reserve (ECR), cost containment reserve (CCR), and price ceiling. For California’s cap-and-trade program, Haiku includes a tax at the price floor level to represent the incentives in the power sector. Renewable Portfolio Standards (RPS) and Clean Energy Standards (CES) are taken from Lawrence Berkeley National Labs 2022 and represented at aggregated regional levels based on a consumption weighted average of state RPS policies. VMM includes a representation of state-level Zero Emissions Vehicle (ZEV) standards. We assume that the EPA waiver upon which these standards depend is reinstated, which represents an important difference from assumptions used in NEMS for generating results in the AEO for the LDV fleet. Gasoline taxes are also represented in VMM at the subnational level.

Clean energy tax credits: New wind generation receives the section 45 Production Tax Credit with new solar generation receiving the section 48 Investment Tax Credit, with both subject to their pre-IRA statutory phasedown and expiration schedules. Clean energy tax credits are modeled with an assumed 13% tax-equity haircut.

Technology assumptions and inputs: For Haiku, Technology costs are sourced from AEO2021 Table 8.2 for the initial calibration process. After calibrating to AEO2021, capital costs for solar, wind, batteries, and new coal and gas CCS plants are replaced with NREL ATB 2022 for all future runs. Electricity consumption for calibration is based on the AEO2021 reference case.

Carbon capture and storage costs are from EPA’s Power Sector Modeling Platform v6 summer 2021 Reference Case, chapter 6. CCS storage availability is assumed to be 90 metric tons in 2030 based on REPEAT’s Preliminary IRA paper, and then doubles every five years thereafter. No currently unplanned CCS plants are permitted to be built in the model prior to 2028.

Emissions and electricity demand for industry, buildings, and non-LDV transportation are taken from the AEO2023 No IRA case.

Central IRA Case

The Central IRA Case includes the policy representations described above for the 2022 reference (with exceptions noted below) along with key provisions from the Inflation Reduction Act as well as EPA’s Good Neighbor rule:

Inflation Reduction Act: Haiku represents modifications to the Section 48 investment tax credit and 45 production tax credits for renewable electricity implemented under the IRA as follows. Optionality for wind and solar generators to choose the PTC or ITC is implemented by calculating the most profitable tax credit (investment based or production based) for a given model plant based on cost characteristics and resource availability. Electricity storage is modeled as receiving the ITC. The production and investment tax credits transition to their respective carbon-intensity based 48E and 45Y credits in 2025.

Tax credit bonuses under the IRA are modeled as follows: For the energy communities bonus, new builds receive a percentage of the 10% or 2.5 cent/kWh bonus varying by each state. The percentage of the bonus credit that applies for a given state is based on the percentage of land area that qualifies according to the treasury guidance. The domestic content bonus is applied uniformly across states as a linear increase to 100 percent of the credit by 2050. Direct pay and transferability provisions are represented as lowering the monetization haircut to 3%.

To represent the IRA’s 45U nuclear production tax credit, the model fixes generation to levels from a baseline scenario without the IRA. 45Q tax credits for carbon capture and sequestration (CCS) are included for new fossil gas and coal plants, with a retrofit option for existing fossil plants. 45Q tax credits provide $85/metric ton for CO₂ that is captured and stored permanently and $65/metric ton for CO₂ captured and utilized, such as for enhanced oil recovery. CCS plants can receive the tax credit for twelve years after the start of operation, and the credit is available for new CCS builds commencing construction no later than 2032.

Credits for the purchase of new electric vehicles are represented in VMM.

EPA’s Good Neighbor Regulation: EPA’s Good Neighbor Regulation is expected to affect both the emissions rates and the capacity of coal plants. To represent the regulation in Haiku, we relied on the modeling performed in EPA’s Good Neighbor Regulatory Impact Analysis (RIA). To represent the effect on emissions rates, we scaled the SO₂ and NOx emissions rates of coal model plants in Haiku to reflect the percent change in state-level average coal emissions rates in the RIA. To represent the effect on capacity, we limited state-level coal capacity of existing coal plants to no greater than the coal capacity projected in EPA’s Good Neighbor RIA.

Emissions and electricity demand for industry, buildings, and non-LDV transportation are taken from the AEO2023 Reference case.

To assess the effects of uncertainty in policy, technology, and fuel availability on outcomes, we explore four sensitivities:

No Good Neighbor

All policies in the 2023 Central Case except for the Good Neighbor regulation.

High Oil and Gas Prices

All policies in the 2023 Central Case with oil and gas prices drawn from AEO2023’s Low Oil and Gas Supply Case scenario. Under this scenario, price paths for oil and gas are higher than in the central case. Emissions and electricity demand for industry, buildings, and non-LDV transportation are also taken from the AEO2023 Low Oil and Gas Supply Case scenario.

Low Oil and Gas Prices

All policies in the 2023 stocktake with oil and gas prices from AEO2023 High Oil and Gas Supply Case scenario. Under this scenario, price paths for oil and gas are lower than in the central case. Emissions and electricity demand for industry, buildings, and non-LDV transportation are also taken from the AEO2023 High Oil and Gas Supply Case scenario.

No CCS Limits

All policies and assumptions in the 2023 central case but with no assumed restrictions on the buildout of CCS transportation and storage infrastructure.

Figure 1 provides a schematic representation of the multi-model configuration used to generate the results in this carbon score.

Haiku operates natively at the resolution of individual states, requiring electricity demand from models operating at other resolutions to be downscaled or aggregated to match. AEO2023 provides outputs of emissions and electricity demand for industry, buildings and transportation (except for LDVs) at the census division level. Electricity demand from these sectors is downscaled from the census division level to the state level using 2021 state-level estimates for input into Haiku. For LDV electricity demand, the VMM provides county level estimates, which are aggregated at the state level for input into Haiku.

Emissions from the power sector are calculated at the model plant level from the model solution and then aggregated to the census division level before being downscaled again to individual power plant locations in EASIUR to calculate health impacts from emissions changes due to policy in the power sector. For LDVs, county level inputs for EASIUR are drawn directly from VMM to estimate the effects of emissions changes from LDVs.

We separate out LDVs from the rest of the transportation by combining the State Energy Data System (SEDS) database for transportation electricity consumption (which excludes EVs) and emissions with the outputs from VMM to calculate the LDV share of the transportation sector emissions and electricity consumption. We hold that these ratios hold constant from 2021 forward for AEO electricity demand and emissions.

To estimate health effects, we first downscale the projected changes in SO₂ and NOx emissions from the census region level to the county level using the National Emissions Inventory, and then calculate air quality changes and health impacts using the EASIUR, described below.

Net present value costs and benefits are calculated using a 2% discount rate. Avoided deaths and benefits from non-GHG pollutants are calculated only using emissions from the electric sector. The social cost of carbon follows Rennert et al. 2022 using the 2% discount rate value. The benefits from non-GHG pollutants are calculated using the avoided deaths from EASIUR multiplied by EPA’s the value of a statistical life adjusted for inflation and income growth where income growth is assumed to follow the growth in average per capita GDP taken from the RFF Socioeconomic Projections (Rennert et al. 2021). Government expenditures only consider the PTC and ITC for clean electricity (sections 45, 48, 45Y and 48E of the tax code) and the tax credit for carbon oxide sequestration (section 45Q of the tax code). The electric vehicle tax credits for new light duty vehicles are modeled in the transportation sector, but the associated government expenditures are not currently tracked and are not included in this analysis.

Model Descriptions

Haiku Electricity Market Model

Model Description: Haiku is a capacity expansion model with 49 nodes for the contiguous states and the District of Columbia. The model distinguishes between competitive and cost-of-service regions and includes region-specific capital and fuel costs. The model is a linear program with perfect foresight over a 31-year time horizon from 2019-2050. It solves for the minimum-cost operation of the electricity system over 24 time blocks representing three seasons, day and night, at four load levels (baseload, shoulder, peak, and superpeak), with a distinct representation of average solar and wind availability for each state and time block. Existing generators are aggregated into up to 18 bins of model plants using data from S&P global. New model plant cost characteristics by technology are sourced from the EIA’s AEO2021 and the National Renewable Energy Laboratory’s Annual Technology Baseline 2022.

The model includes build constraints to capture the effects of historical barriers to infrastructure deployment. Capacity for a given fuel type is not allowed to grow faster than 7 percent per year starting in 2023. Carbon storage and transportation and capacity are set based on EPA’s Power Sector Modeling Platform v6 with total annual storage availability scaled to 91 million metric tons in 2030, doubling each year thereafter based on assumptions in Princeton’s REPEAT IRA Preliminary Report August 2022.

Haiku is designed to represent policy changes from the national baselines set by EIA in the Annual Energy Outlook (AEO). For electricity demand, state level demand for residential, commercial, industrial, transportation, and electric vehicles are calculated from AEO2021 Electricity Market Module (EMM) region level outputs. State level generation outputs from EIA are also imputed from AEO2021 EMM regions to calculate the total generation a state is projected to have. In a calibration stage, constraints require the model solution to be within 1% of several values using these imputed values from AEO:

• 47 of 49 (to allow for feasibility in calibration) states’ total generation
• National levels of generation of each fuel type

The shadow prices from these constraints are applied to all future scenario runs.

Vehicle Market Model (VMM)

Model description: The RFF Vehicle Market Model is a model of the US market for new passenger vehicles and light trucks. Underpinning the model is a unique dataset of 1.5 million vehicle purchase decisions made by households during 2010-2018 and updated on a regular basis. These data include a detailed description of the vehicle purchased, other vehicles considered for purchase, and household demographics and economic characteristics. Consumer preferences for vehicle attributes are estimated from the survey data. The model integrates household preferences for vehicle attributes with manufacturer decisions of vehicle technology and pricing and of entry of new electric vehicles. This model can simulate the economic and environmental impacts of policies that affect the new vehicle market. Such policies include, but are not limited to, fuel economy and greenhouse gas standards, electric vehicle mandates and subsidies, electric vehicle charging infrastructure investments, gasoline and/or carbon taxes, and greenhouse gas cap and trade programs. Extensions of the model to the region and state level enable analysis of policies such as the zero-emissions vehicle mandate, regional carbon policies, state gasoline taxes, and electric vehicle subsidies (including charging infrastructure investments).

VMM outputs vehicle emissions and fuel demand at the county level. The model solves from 2018-2030. After 2030 we project the difference between AEO2023 and the VMM outputs (emissions and electricity demand) by census division in 2030 and assume that difference holds out to 2050. The census division differences are distributed by state based on the state share of each output in 2030 before being input into Haiku.

Estimating Air pollution Social Impact Using Regression (EASIUR)

Model Description: EASIUR (Estimating Air Pollution Social Impact Using Regression; Heo et al. 2016, 2017) employs a regression model based on the CAMx chemical transport model to gauge the public health costs linked to premature mortality due to primary and secondary PM2.5 emissions in the United States. The model calculates marginal damages in USD per metric ton for four different species (EC, SO₂, NOx, and NH3) across four seasons and the annual average of these seasons. The analysis is conducted at a spatial resolution of 36 km x 36 km, considering three stack heights: ground level area emissions, point source emissions at 150 m, and point source emissions at 300 m.

Utilizing this comprehensive dataset, we generate a county-by-county annual average source-receptor matrix for individual pollutants, assuming the emission source is located at the county’s centroid. Social damages are then computed for over 3000 counties in the US, evaluating the impact of emissions at the source county. This calculation is repeated for all individual source counties, pollutants, and stack elevations. The emissions from state electricity generation, categorized by pollutants and by fuel types, are downscaled to the county level emissions using factors derived from the 2017 National Emission Inventory. Through the integration of county level emissions data and the county level source-receptor matrix, the analysis enables the estimation of public health costs associated with all county receptors. Estimates of the changes in health outcomes are based on changes in premature mortality using Krewski et al. (2009) and valued at $8.8 million in 2010$ and 2010 income adjusted for income growth using an elasticity of 1 where income growth is based on the mean projected gdp per capita in the RFF SPs. Subsequently, the county-level public health costs are aggregated to provide results at the state and national level.

Policy Assumptions

For this analysis we do not distribute changes in government revenue or expenditure directly to households. Government expenditures lower retail prices and generator profits by lowering the cost of generation reflected in the wholesale prices. Government expenditures also raise generator profits for clean generators. We assume that PTCs affect retail prices and producer profits within the ten year time window that a generator is eligible for them. We assume that ITCs affect retail prices and producer profits in a way that in annualized over the lifetime of the asset.

References

NEMS Documentation
VMM Documentation
Annual Energy Outlook 2021 - U.S. Energy Information Administration (EIA)
Annual Energy Outlook 2023 - U.S. Energy Information Administration (EIA)