Decarbonization pathways for the residential sector in the United States

Residential GHG emissions in the United States are driven in part by a housing stock where onsite fossil combustion is common, home sizes are large by international standards, energy efficiency potential is large and electricity generation in many regions is GHG intensive. In this analysis, we assess decarbonization pathways for the US residential sector to 2060, through 108 scenarios describing housing stock evolution, new housing characteristics, renovation levels and clean electricity. The lowest emission pathways involve very rapid decarbonization of electricity supply alongside extensive renovations to existing homes, including improving thermal envelopes and heat pump electrification of heating. Reducing the size and increasing the electrification of new homes provide further emission cuts and combining all strategies enables reductions of 91% between 2020 and 2050. The potential of individual mitigation strategies shows great regional variation. Reaching zero emissions will require simultaneous deployment of multiple strategies and greater reduction of embodied emissions.

R apidly reducing GHG emissions from buildings is central to mitigating global climate change. The United States has one of the highest levels of per capita residential energy use in the world (1.5 times the Organisation for Economic Co-operation and Development (OECD) average) and the second largest total residential energy use 1 . Recent reductions of residential energy-related emissions in the United States have primarily derived from decarbonizing electricity supply, with much smaller reductions from energy efficiency and increased use of electricity for space heating 2 . Although complete electrification of residential energy is feasible, complete decarbonization of electricity supply is challenging 3,4 . Reducing residential energy demand through efficiency and sufficiency can alleviate this challenge. Two approaches to improve building stock energy efficiency are renovating existing buildings and replacing older stock with new buildings [5][6][7][8] . Comparing these approaches requires consideration of emissions 'embodied' in construction, which constitute around 9% of residential emissions in the United States 9 . Literature has not converged on which approach is preferable 10 but existing comparisons of renovating and replacing usually focus on individual buildings or neighbourhoods; rarely have they been made for building stocks of an entire country 8 .
'Sufficiency' approaches to reducing GHG emissions are a recent addition to climate change mitigation discourse and target reduced demand for energy and materials, while delivering well-being for all 11,12 . For residential buildings, sufficiency can be translated into a global convergence of floor area per person 13,14 of somewhere in the range of 15-40 m 2 per capita (refs. [15][16][17][18]. Current average floor area usage in the United States is 60 m 2 per capita (ref. 19 ), one of the highest levels globally 20 , although there is considerable variation within the United States in floor area consumption by house type, geography and race 19,21 .
In this paper, we estimate emission pathways from operation, construction and renovation of residential buildings in the United States in 108 scenarios from 2020 to 2060. The primary aim is to assess potential GHG emission reductions from individual and combined strategies applied to existing homes, new homes and electricity supply; and to illustrate how these potentials vary regionally by climate, housing stock characteristics, electricity grid region and population growth. The consideration of embodied emissions, engineering-based energy modelling using high-resolution housing characteristics and representation of detailed renovation measures based on empirical renovation data are new aspects of this work. Results show that deep renovations of existing homes and rapid decarbonization of electricity supply have the greatest potential for emission reductions. Reducing the size of the largest new homes and increasing the electrification and multifamily share of new housing can deliver substantial further reductions, but a faster replacement of existing homes does not reduce emissions.

Description of scenarios
The 108 emissions scenarios (3 × 4 × 3 × 3) are developed, defined by three scenarios describing evolution of the United States housing stock 19 , four scenarios describing characteristics of new housing, three renovation scenarios and three electricity supply scenarios 22,23 ( Table 1). The scenarios incorporate different approaches to climate change mitigation; sufficiency approaches are represented in the high multifamily growth and reduced floor area scenarios, efficiency improvements occur in renovation, high stock turnover and increased electrification scenarios, while energy supply decarbonization is represented in electricity supply scenarios. In the higher ambition scenarios, historic trends are altered to levels that are optimistic but still feasible; they do not necessarily achieve the maximum technical potential. For instance, the increase in renovation rates by factor 1.5 reflects the possibility for increased renovation rates but also practical constraints on how much they can increase. The scenarios are not optimized to achieve a specific emissions reduction target. Further descriptions of the scenarios, modelling approach and limitations, are provided in the Methods.

Assessment of climate change mitigation strategies
Annual GHG emissions decline in all scenarios, and the extent of decline is largely explained by the extent of electricity decarbonization and renovation depth (Fig. 1). In 2030, 55 out of 108 scenarios meet a US government target for reducing emissions by 50% compared to 2005 23 ; this generally requires at least low renewable energy cost electricity (LREC) and advanced renovation (AR). Only one scenario reduces emissions by 50% between 2020 and 2030, which is a global reduction required to limit climate change to 1.5 °C warming 24 . This scenario has carbon-free electricity by 2035 (CFE), extensive renovation (ER), smaller (reduced floor area, RFA) new housing, increased electrification (IE) and high multifamily (high-MF) stock growth. In 2050, 36 scenarios reduce emissions by at least 80% relative to 2005, which was the 'mid-century strategy' outlined in the US nationally determined contribution to the Paris 2015 agreement 25 . Meeting this target requires either LREC electricity combined with ER, CFE with AR/ER, or CFE with regular renovation (RR), lower floor space (RFA and high-MF) and IE of new housing. Due to residual emissions from residential fossil fuels and construction, even the most ambitious scenarios modelled do not meet a 1.5 °C-consistent goal of zero emissions in 2050 24 . Cumulative 2020-2060 emissions range from 12.0 to 28.9 GtCO 2 e (Fig. 2). A current-population-based allocation of the remaining global carbon budget to meet 1.5 °C with 50% likelihood 26 gives the United States a budget of 21 GtCO 2 e from all sources from 2020. This would be even smaller under fair effort-sharing allocations 27 .
Within each electricity supply and renovation scenario set, housing stock evolution and the characteristics of new housing provide further variations in emissions. Annual emissions are lower by 33-54 MtCO 2 e yr −1 in 2050 if building new homes with RFA and IE compared to baseline new housing characteristics, while cumulative 2020-2060 emissions see reductions of 1.1-1.8 GtCO 2 e (Fig. 2). With mid-case (MC) electricity, RFA has greater potential for emission reductions than does IE but, if electricity supply decarbonizes completely by 2035, these two strategies have about the same cumulative potential (Fig. 2c). These strategies are complementary, so the largest emission reductions occur when they are combined. Due to higher embodied emissions 19 and the size difference between old and new single-family housing (Extended Data Fig. 1), high-turnover (high-TO) stock results in increased emissions

Regular renovation (RR) (Supplementary Section 3)
Renovation continues at historic rates, moderate efficiency improvements and slow electrification of space/water heating.
Advanced renovation (AR) Renovation rates increase by factor of 1.5 relative to historic levels (leading to earlier retirements of existing equipment), with higher-efficiency improvements and moderate increase in electric share of space/ water heating equipment replacements.
Extensive renovation (ER) Similar to AR except higher share of heat pumps in space/water heating renovations. All fossil space heating equipment replaced with electric heat pumps from 2025.  ( Fig. 2), despite improvements in efficiency which accompany faster growth of new housing. Therefore, renovating has far greater emission reductions potential than does replacing existing homes. Like RFA, high-MF stock growth reduces both embodied emissions and future energy-related emissions. High-MF stock growth reduces cumulative 2020-2060 emissions by 0.30-0.81 GtCO 2 e compared to baseline stock growth (Fig. 2). While extensive renovation and rapid electricity decarbonization both have high potential, there are limitations to relying on either strategy individually, as illustrated by Figs. 3a,b and 4a,b. With ER and gradual (MC) decarbonization of electricity (Fig. 3a), emissions from fossil combustion decline markedly but emissions from electricity remain substantial. Conversely, with faster (LREC) electricity decarbonization but RR of existing homes (Fig. 3b), emissions from fuel use remain large and are locked in for decades (beyond 2060) through installation of fossil-based replacement heating equipment. The most impressive emission reductions result from the combination of rapid (CFE) decarbonization of electricity and ER, with further reductions possible from construction of new homes that are smaller, electrified and more multifamily (Extended Data Fig. 2). With the most optimistic combination of renovation, new housing and stock scenario dimensions, the additional emission reductions from completely decarbonizing electricity by 2035 (CFE) compared to LREC are immense: 2050 annual emissions reduce from 227 to 83 MtCO 2 e yr −1 (Fig. 3c versus d) and cumulative 2020-2060 emissions reduce from 17.7 to 12.0 GtCO 2 e (Fig. 2). After electricity emissions reach zero in 2035, subsequent reductions in CFE scenarios are more gradual and annual emissions decline slowly from 2045 onwards ( Figs. 1 and 3d). In CFE-ER scenarios, most emissions from 2050 onwards are from construction (Fig. 3d). Embodied emissions projections assume improvements in material production and construction activities leading to ~23% reductions in average embodied emission intensities (kgCO 2 e m −2 ) by 2060 19 . Greater reductions in embodied emissions could result from building without basements or garages, substituting wood for concrete structural elements, using low-carbon cementitious materials, greater electrification of construction transport and energy use and avoiding insulation with high-global warming potential blowing agents 19,[28][29][30] .

Electricity supply scenarios
Our lowest emission scenario shows combined energy and embodied emissions of 0.25 tCO 2 e per capita by 2050, down from 2.74 tCO 2 e per capita in 2020. This is lower than 2050 per capita US residential emissions in the lowest emission scenario from ref. 31 (0.62 tCO2 2 e per capita, energy emissions only) but higher than 0.15 tCO 2 e per capita from the lowest emission scenario by ref. 17 (embodied emissions and energy emissions from heating, cooling and hot water only) or 0.17 tCO 2 e per capita from the International Energy Agency's (IEA) sustainable development scenario 32 (energy emissions only).

Geographical variation in strategy potential
Tremendous geographical variation exists in the effectiveness of strategies, depending on local housing stock characteristics, GHG intensity of electricity, population projections and climate. Figure 4 compares percentage reductions in cumulative 2020-2060 emissions by state between ER, LREC electricity, IE-RFA new housing and high-MF stock growth strategies. CFE electricity is excluded from this comparison, as LREC represents a less challenging yet still optimistic electricity supply scenario to compare against non-electricity strategies.
ER has greatest influence in regions with cold/mixed climates, low GHG-intensity electricity, low shares of electric heating, New housing characteristics  Table 1); columns show variation by housing stock evolution and renovation scenarios (abbreviations as in Table 1). *Identifies scenarios that meet the 2050 target of 20% of 2005 emissions or lower.
growth and large differences in electrification between single-and multifamily homes. For both IE-RFA and high-MF, relative emission reductions are greatest in California. Absolute emission reductions from each strategy are largest in populous states like Texas, New York and California ( Supplementary Fig 30). The most effective strategy can be identified in each county and is most often ER or LREC (Extended Data Fig. 5a). Excluding electricity supply scenarios shows that IE-RFA and high-MF can be preferable to ER in fast-growing counties in states including Texas, Florida and Georgia (Extended Data Fig. 5b).

technical, economic and policy challenges to mitigation
Technical challenges for renewable-driven electricity decarbonization include diurnal and seasonal balancing of supply and demand and maintaining grid stability with high penetration of inverter-based (wind and solar) technologies 4 . The MC/LREC scenarios 22 project factor 4.8/6 increases in combined wind and solar generating capacity between 2020 and 2050, requiring average annual combined wind and solar increases of 26 or 35 GW, respectively, with growth in solar including increases in residential rooftop higher shares of old housing and lower population growth. New England (northeast United States) and New York state demonstrate the greatest potential, with 31-35% reduction of cumulative emissions (Fig. 4a). LREC electricity supply has the greatest influence in regions with high 2020 electricity GHG intensity, greater reductions in electricity GHG intensity (Extended Data Figs. 3 and 4) and high shares of electric heating. In relative terms, the greatest reductions occur in Missouri, Kansas, Tennessee and Kentucky (30-34%; Fig. 4b). Combining ER and LREC provides high reductions in regions that benefit from each strategy individually and especially large reductions in regions with colder climates, low electric heat share and older homes but relatively GHG-intensive electricity in 2020 ( Supplementary Fig 29b). One illustrative example is Wisconsin, where reductions from ER and LREC individually are 7% and 20%, respectively, but the combined reduction from ER-LREC is 38%.
Constructing new homes with IE and RFA has greatest influence in areas with less GHG-intensive electricity, high population growth and lower shares of electric heating, while emission reductions from high-MF stock growth are greatest in regions with high population  envelope upgrades improve 7 million housing units per year by 2040, while 6-7 million heat pumps will be installed in existing homes from 2035 onwards (Extended Data Fig. 6). Including installations in new homes, annual demand for heat pump units could grow to 9 million in 2050. Such growth in heat pump supply appears possible considering current growth trajectories ( Supplementary Fig. 12) and industry capacity for producing air-conditioning units, which have similar manufacturing requirements 36 . Fossil fuel to heat pump replacements offer substantial GHG reductions, especially in cold climates; these replacements are economical when replacing fuel oil or propane but can lead to higher costs when replacing natural gas equipment ( Supplementary Fig. 17). This is one potential economic barrier to residential decarbonization. Targeted policy support may therefore be required for gas to heat pump renovations. Envelope renovations offer high GHG reductions, particularly in cold regions and in homes without much insulation, and are usually economic ( Supplementary Fig 19). Combined heating system and envelope renovations offer the largest emission reductions, particularly for fossil to heat pump replacements in cold climates ( Supplementary  Fig. 20). Energy reductions from envelope and heating renovations are largest in older (<1960) single-family homes in cold climates (Supplementary Tables 6 and 7). Our economic assessment of renovation strategies considers only renovations occurring through 2025 and is subject to considerable uncertainty surrounding future equipment and energy costs and discount rates. Residential renovations in the United States are supported by a patchwork of utility, federal and local initiatives, including utility efficiency programmes, low-income weatherization programmes and tax credits. Federal standards set minimum efficiency levels for replacement equipment but the levels have historically been photovoltaic (PV) systems. This is comparable to current growth rates. Between 2019 and 2020, combined wind and solar capacity grew by 30 GW (ref. 33 ), up from 7 GW growth between 2010 and 2011. A continuation of growth rates since 2014 would see annual increases of 102 GW by 2030. The closest description of an electricity system comparable to the CFE scenario is the 2050 100% renewable electricity scenario from ref. 3 . To reach 100% renewable supply by 2035 would require average annual increases in combined wind and solar capacity of 119 GW from 2020. Assuming no land requirements for offshore wind and distributed rooftop PV, land use for wind and solar would grow from 41,800 km 2 in 2020 to 92,000 or 152,800 km 2 by 2050 in MC or LREC, respectively, or 179,300 km 2 for CFE. Electricity grids approaching 100% renewable generation may exhibit nonlinear increases in incremental system costs above 95% renewable generation 3 . Increased transmission connection between Eastern and Western Interconnections in the United States can reduce costs of electricity supply and enable lowest-cost generation mixes with high (85%) renewable penetration 34 . Increased transmission and electricity storage capacity can also help to smooth regional and temporal imbalances in electricity demand and supply and will be important in energy systems with increased end-use electrification and high penetration of variable renewable generation 35 . For seasonal supply-demand imbalances, alternative storage solutions such as power-to-hydrogen may be required 4 .
ER of existing homes requires increased insulation, reduced infiltration and replacement of space and water heating equipment with high-efficiency heat pumps and electric water heaters. Supplementary Section 3 details the changes in equipment and envelope characteristics in renovation scenarios, as well as costs and emission reductions from specific renovation measures. With ER, most from rapid electricity decarbonization. Our least-emission scenario still projects 12 Gt of cumulative CO 2 e emissions between 2020 and 2060, which is 56% of the United States's entire carbon budget for meeting 1.5 °C with 50% likelihood. Targeting 1.5 °C would therefore require solutions beyond the most ambitious scenarios presented here, including more comprehensive reductions of embodied emissions, through restrictions on floor area growth and innovations in material production.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41558-022-01429-y. set separately for electric and fossil equipment and thus have not encouraged adoption of more efficient electric equipment over fossil alternatives 37 . Very few state or utility efficiency programmes reward energy or emissions savings from fuel switching 38 , while numerous states discourage or prohibit fuel switching 39 . The Better Energy, Emissions and Equity initiative 40 launched in May 2021 aims to accelerate the adoption of heat pump water heaters and improve the performance of cold climate heat pumps. The proposed Build Back Better Act also includes rebates for qualifying electrification projects, which could boost heat pump replacements. Heat pump electrification of space heating can be more cost-effective when purchasing new or replacement air conditioners, so that heating and cooling equipment costs are combined 36 . This is not considered in our economic assessment of renovations but could substantially improve the economics of natural gas to heat pump renovations. Although housing tenure is outside of our model framework, split incentives between landlords and tenants are another potential barrier to residential renovations 41 . To address this, efficiency programmes can target rental homes with incentives, alternative financing solutions (for example, on-bill financing) and building performance standards to ensure accelerated renovations of rental housing, particularly in communities with lower access to clean efficient energy 21 .
Challenges surrounding building fewer large homes or more multifamily homes mostly relate to policy and societal norms. Policy options include introducing size limits, removing zoning and other local restrictions on denser housing 42,43 and restructuring Federal tax policies which make multifamily investments costlier than single-family 44 . RFA and high-MF stock growth could stabilize floor area per capita at current levels 19 but will not induce substantial reductions (Extended Data Fig. 7). Thus, reducing floor area consumption to sufficiency levels (40 m 2 per capita or lower) cannot be done by focusing on new housing alone 19 ; doing so would require strategies for existing homes, such as household sharing 45 or converting existing single-family homes into multiple housing units. In our lowest emission scenarios, where construction becomes the majority emission source by mid-century, such measures targeting existing homes would reduce the need for new construction and make zero emissions targets much more attainable. Otherwise, the elimination of embodied emissions will rely on material efficiency 17 and greater advances in low-carbon material selection and production 28,29 .

Conclusion
In this paper, we assess decarbonization pathways for residential buildings in the United States in 108 scenarios to the year 2060, incorporating embodied and energy-related emissions. Our analysis delivers new insights into how much emissions can be reduced from different mitigation measures, in various segments of the housing stock. The pathways with lowest emissions require rapid decarbonization of electricity alongside extensive electrification-focused renovations of existing homes. Most of the energy-related emissions in 2060 will be from homes that exist today. However, increasing the turnover of housing stock will not reduce residential emissions due to higher embodied emissions and because efficiency benefits are partially offset by larger new homes. Accelerated and deep renovation of existing homes is therefore a crucial component of residential decarbonization. Envelope and heating renovations are particularly impactful in regions with cold climates, low shares of electric heating and large shares of old homes. In new homes, substantial emission reductions arise from avoiding construction of excessively large houses or increasing the electrification of heating, especially when combined with rapid grid decarbonization. The characteristics of new homes have greatest influence on emissions pathways in regions with strong population growth. Regions with high shares of electric heating and high GHG intensity of electricity benefit higher shift towards electric space and water heating systems and heat pumps in particular. In the ER scenario, much higher rates of electrification of space and water heating take place, with 100% of replacements of fossil space heating renovations with electric heat pumps and 100% electrification of fossil water heating equipment from 2025 on. This does not mean that all existing fossil space heating equipment is replaced by heat pumps in 2025 but, if a fossil-based space heating system is replaced, it is replaced by a heat pump . Tables and figures  showing the assumptions and results of the renovation scenarios are presented in  the Supplementary Section 3. We calculate energy-related GHG emissions using standard emission factors for combustion of fossil fuels 55 and annual average CO 2 intensity for three electricity supply scenarios: MC and LREC from the National Renewable Energy Laboratory (NREL) standard scenarios 22 and a scenario involving 100% carbon-free electricity by 2035 (CFE) 23 . The MC is the baseline electricity supply scenario, while LREC is the NREL standard scenario with the fastest decline of electricity GHG intensities. For CFE, in the absence of data describing regional projected electricity generation by source, we assume the same intensities as LREC until 2025, which then halve between 2025 and 2030, before reaching zero by 2035. Weighted average electricity GHG intensities are calculated at the level of 20 generation and emission assessment (GEA) regions 46 (Extended Data Fig. 4), which are defined to approximately match the US Environmental Protection Agency (EPA) eGRID regions. The GEA regions are aggregations of the smaller 134 balancing areas used to define future electricity generation and consumption. Aggregation to GEA region was preferred to using the higher resolution balancing area GHG intensities, as the intensities for individual balancing areas will fluctuate a lot as a result of electricity trading, whereas the eGRID (and GEA) regions are designed to be more reflective of average electricity grid characteristics in larger areas, with reduced influence of electricity trading. Energy-related emission intensities describe CO 2 emissions from combustion only 22 , excluding upstream emissions such as fugitive methane releases from fossil fuel extraction or embodied emissions from electricity generation and transmission infrastructure. Residential fossil combustion includes non-CO 2 combustion products but CO 2 emissions account for over 99% of total combustion GHGs 55 . Embodied emissions from material production and construction 19 are based on material life cycle assessment databases, environmental product declarations and literature and include non-CO 2 GHGs. To estimate the land requirements of wind and solar generation in the electricity supply scenarios 22 , we divide the generating capacity of onshore wind, utility PV and distributed PV 56 by technology-average installed capacity density coefficients for renewable electricity in the United States. On the basis of available literature we use capacity densities of 3 W m −2 for onshore wind 57 , 50 W m −2 for utility PV 58 and 25 W m −2 for concentrated solar power 59 . Land requirements for offshore wind and distributed PV (which largely corresponds to rooftop PV) are assumed to be zero. Without information on types or quantities of biomass feedstocks used for bio-electricity generation (which is negligible in MC and LREC but shows notable growth in later years of CFE), land use for growing feedstocks for biomass electricity is not considered.
Growth of residential rooftop solar PV is incorporated in the electricity supply scenarios via the distributed generation market demand model 60 , not on the demand side through residential renovations. As growth of residential PV is already reflected in the electricity supply scenarios, we do not consider it as an additional residential renovation measure, to avoid double-counting.
Energy simulation. Calculation of energy consumption in the US housing stock is performed using ResStock, a residential energy simulation tool with high-resolution characterization of the US housing stock. Built on the OpenStudio/ EnergyPlus building energy simulation engine, ResStock draws on an extremely rich description of US residential building characteristics at various geographical resolutions ranging from national to county and PUMA depending on the characteristic in question 48,61 . County-specific weather files are used to reflect local climate and simulations are made over a representative year (TMY3) at a 10-min resolution. Changes in weather files due to climate change are not incorporated. Variation in energy demand for electronic appliances by Census Division and house type is represented but we do not simulate future changes in this energy demand segment, although it could grow in line with increasing number and size of personal electronics per household 2 . Such growth could be balanced by decreased TV ownership and increased appliance efficiency. Housing stocks in Hawaii and Alaska are not included in ResStock (or the analysis presented here) due to limited availability of housing characteristics data in these states. We do not include current or future energy consumption in vacant housing units in this analysis.
Energy simulations representing the entire contiguous US housing stock are made for the year 2020 and for every 5 years between 2025 and 2060, for each housing stock, new housing characteristics and renovation scenario combination. Energy-related GHG emissions are calculated on the basis of energy consumption by energy carrier in each year and are interpolated for the intervening years in which energy demand is not simulated (for example, 2021-2024) using the spline() function in R. To capture the heterogenous characteristics of the US housing stock in a representative manner 49 , we simulate energy consumption in many houses for each scenario and simulation year, so that one simulation represents somewhere in methods Housing stock and characteristics scenarios. Housing stock evolution and new housing characteristics scenarios are based on scenarios developed by Berrill and Hertwich 19 , which is the source of estimated embodied emissions from material production and construction and where a full description of the housing stock model can be found. County population projections 47 drive the housing stock model 19 and are scaled to the mid-range scenario from the 2017 US Census Bureau population projections to 2060 48 . The scenarios are extended for this work to include the scenario of increased electrification in new housing and to describe scenarios renovation of existing housing.
Emissions from material production and onsite energy and transport in new construction are calculated for 51 housing archetypes 19 , capitalizing on high-resolution representation of US housing characteristics by house type, size, foundation type, heights and so on in the ResStock housing characteristics data 49 . Embodied emissions from renovation activities are included for envelope renovations only. For a given archetype, an envelope renovation is assumed to require 10% of the cement, gypsum, glass and wood products and 70% of the insulation materials required for an equivalent new construction 19 . Embodied emissions from energy equipment such as furnaces and heat pumps are not considered. Our calculations of embodied emissions from construction 19 incorporate moderately optimistic assumptions on reduction in GHG intensity of material production by 10-50% between 2020 and 2060, depending on the material 50 . This reduces emissions per m 2 floor space by on average 23%.
The new housing characteristics scenarios are implemented by altering the ResStock housing characteristics data for new housing cohorts before generating a representative sample of new housing built in eight 5-year periods spanning 2021-2060 (2021-2025, 2026-2030 and so on). Future housing characteristics are modified, depending on anticipated adoption of residential building energy codes by states 51 , updates to federal energy appliance standards 52 and assumptions on increased electrification and efficiency improvement of equipment and insulation. Building energy codes mostly apply to building envelope characteristics, such as insulation and infiltration levels and energy ratings of windows 53 , while the federal efficiency standards apply to energy-consuming equipment and appliances, such as space and water heaters, air-conditioning systems and refrigerators. We also incorporate assumptions regarding changes and trends in housing and energy appliance characteristics that are not directly based on codes and standards, such increased adoption of electric equipment used for space and water heating, increased use of heat pumps and continued growth of air-conditioning equipment ownership.
In the baseline new housing characteristics scenario (A), housing built in the next four decades has the same regionally specific characteristics as housing built in the 2010s. The exception is fuel choice for space and water heating, cooking and clothes drying, where we assume electricity to be a more common choice in new housing and the electricity share to increase every decade ( Supplementary Fig. 22).
In the RFA scenario (B), no new housing unit exceeds 279 m 2 (3,000 ft 2 ), an arbitrary limit chosen on the basis of the floor area bins used in the ResStock housing characteristics database and originating from the American Housing Survey 54 . Housing units that previously fit into the two largest size categories of 279-371 m 2 (3,000-3,999 ft 2 ) or 372+ m 2 (4,000+ ft 2 ) are reassigned to be in one of the 186-232 m 2 (2,000-2,499 ft 2 ) or 232-279 m 2 (2,500-2,499 ft 2 ) categories with 50:50 probability (Supplementary Fig. 25). In the IE scenario (C), electrification of new housing is much faster, with all regions reaching complete electrification by 2030 except the northeast, which is fully electric by 2040 ( Supplementary Fig. 22). The IE and RFA scenario (D) simply combines the new housing characteristics scenarios B and C. Further information on new housing characteristics scenarios is provided in Supplementary Section 4.
Renovation and electricity supply scenarios. Our analysis represents the most comprehensive existing assessment of the emission reductions from residential retrofits over the coming decades, incorporating energy-relevant characteristics of existing housing units up to the county and public use microdata area level and the most recent empirical data on recent renovation trends, and estimating energy reductions of renovation actions with a detailed engineering-based simulation. We consider energy-related renovations applying to addition/replacement of space heating, space cooling and water heating equipment and envelope upgrades for crawlspaces, unfinished basements, external walls and unfinished attics, which increase the R-value of those building assemblies and reduce the infiltration of the building envelope. These measures capture the main types of renovations that offer substantial potential for energy reductions 49 . Two pieces of information are required for each renovation, the rate of renovation in the housing stock (the probability of a housing unit making a specific type of renovation in a given year) and the characteristics of a given system postrenovation, conditional on its prerenovation status.
We define three renovation scenarios: 'regular' , 'advanced' and 'extensive' . The RR scenario is based on a continuation of recent trends, a moderately optimistic implementation of the depth of renovations and low-moderate rates of replacing fossil heating equipment with electric alternatives. In the AR scenario, we multiply the probability of undergoing renovations by a factor of 1.5 and we give stronger preference to higher efficiency replacements, including a moderately future emission trajectories, such as the rate and depth of renovations and decarbonization of electricity supply. Combining the selected values for each varying input parameter created 108 unique scenarios ( Table 1). The range of emissions trajectories demonstrated by these scenarios are not intended to represent all possible future emission pathways. Parameter values excluded from our scenarios space, which would probably result in notable differences to emission pathways estimated, include higher or lower population and housing stock growth trajectories, slower decarbonization of electricity, slower renovation rates and increased growth in size of new single-family housing. A rather pessimistic scenario, assuming fixed electricity GHG intensity at 2020 levels and no renovation of existing housing, is included in our illustration of annual emissions 2020-2060 in Extended Data Fig. 2. This can be considered as a worst-case outcome for future emissions and shows almost no change in the level of annual emissions over the next 40 years. Other measures to reduce residential energy and emissions were excluded from our analysis. These include behavioural changes 72 and reduction of per capita floor space in existing homes through household sharing (increased household size) or subdividing large homes to multiple smaller units.
For embodied emissions, faster reductions in the GHG intensity of construction could result from greater technological advances in the production of high-emitting materials such as cement, steel and insulation products, increased use of lower-carbon materials in construction 28 and low-carbon electrification of construction site energy use and transport. A faster decarbonization of construction activity could alter our current conclusions on increased emissions from faster housing stock turnover. However, the finding of much greater emission reduction potential from renovation of existing housing, compared to faster rebuilding, would not be changed.
Annual average electricity emission intensities were used instead of short-run or long-run marginal emission rates, as the annual average intensities are more appropriate for very large changes in electricity use across the entire residential sector 56 . Using long-run marginal emission rates would be more suitable for quantifying the emission impacts of incremental changes to the housing stock or individual renovations. In broad terms, average emission intensities happen to be similar in magnitude to long-run marginal rates, so we would not expect major differences if we were to use long-run marginal rates instead of the averages intensities used in this analysis. Using hourly emission rates may be more suitable when considering the time of day of residential electricity demand vis-à-vis electricity demand from other sectors but was outside the scope of the present analysis. Further, complex interactions can exist between energy efficiency measures and demand response strategies such as peak-load shedding and load shifting-which provide additional benefits to electricity grids, such as flexibility and reduced peak generation capacity and can facilitate higher levels of variable renewable generation 73 . Such temporal considerations and intersectoral interactions were outside of the scope and represent a promising avenue for future research considering increased electrification in all sectors 35 .
Costs and benefits of residential renovations extend beyond the capital and energy expenses considered here. Substantial human health benefits have been demonstrated from changes in indoor and outdoor air quality associated with building efficiency improvements and fossil fuel to electricity fuel switching 74,75 . Additional health and mortality benefits, associated with reduced exposure to excessively low and high temperatures, can be expected from improvements in envelope efficiency 76,77 . Quantification of these health impacts was beyond the capability of our model; our cost-benefit analysis of renovations therefore does not incorporate health-related costs or benefits.
While a cost-benefit analysis could have been applied to other families of mitigation measures, we limit the cost-benefit analysis to the renovation strategies. In the electricity supply scenarios, costs and benefits would result from changes in electricity prices and changes in GHG and other environmental emissions. Due to model input assumptions in the dispatch model which generated the MC and LREC scenarios, electricity prices are lower by around 10-15% from 2030 onwards in LREC, compared to MC 56 . Our CFE electricity scenario is not generated by an electricity dispatch model, although the 100% renewable electricity scenarios generated by ref. 3 are conceptually similar. The system cost estimates provided by these models represent power system costs including costs of building and retiring capital assets as well as energy costs but exclude transmission maintenance, distribution and administration costs 3,46 . These costs are not designed to reflect retail electricity rates and thus a cost comparison of electricity supply scenarios for end-use consumers is not feasible. There are numerous costs and benefits associated with other scenario dimensions (high-MF, RFA, IE and high-TO) as these would influence transportation patterns, housing costs, access to urban amenities, privacy and so on. These are highly dependent on location and individual preferences and quantification of these costs and benefits was excluded.
Exclusion of housing tenure overlooks the possibility for higher energy consumption and lower propensity to invest in energy efficiency renovations in rental housing. We investigate this issue further in Supplementary Section 6.3. Demographic variables such as household income, race and ethnicity are also excluded from the housing stock model and the energy simulation model. As such, assessment of access to energy efficiency renovations by population groups and estimation of distributional effects of the various emission reduction strategies were not possible within our current modelling framework. the range of 590-800 homes. A total of 180,000 simulations are used to represent the 2020 occupied housing stock of 122,516,868 homes. In all, 3.412 million building simulations are used to represent the complete set of scenarios. For each simulation, the weighting factor (how many homes are represented) is modified over the projection period to reflect the loss of housing of a given type, cohort and county combination from the occupied housing stock, based on the housing stock model outputs 19 . Figure 1 summarizes the data inputs, assumptions and various components of the model, which produces outputs of annual energy consumption by end-use and fuel, GHG emissions associated with energy use and material flows and GHG from new construction, for housing stocks by type and cohort in each county. As ResStock does not contain data for Alaska and Hawaii, our scenario results apply to the contiguous United States, where 99% of national energy-related GHG emissions occur 62 . As a basis for the 2030 and 2050 emission reduction targets indicated in Fig. 1, we calculate total residential emissions in 2005 and 2020 by combining residential energy emissions 63 with emissions from construction of new housing in 2005 and 2020, scaled by 0.99 to exclude Alaska and Hawaii. Historical emissions from construction are calculated by multiplying numbers of single-family and multifamily housing units completed 64 and manufactured housing shipments 65 , by year-and type-specific average house floor area 65,66 , by the average embodied GHG intensities per unit floor area of each house type 19 .

Model integration. Supplementary
Enviro-economic assessment of renovation strategies. We compare costs and benefits (private economic costs/savings, GHG reductions) for detailed renovation measures that take place between 2021 and 2025 over a 25-year time horizon (2026-2050), to assess their emission reduction potential and economic feasibility. Only private costs and benefits were quantified in economic terms; no societal costs or benefits (for example, related to air quality, health impacts and economic damages from GHG emissions) were quantified. Net present value (NPV) and abatement costs should be interpreted accordingly as the private economic value of investments in efficiency and energy equipment. They do not reflect the socially optimal performance of efficiency investments which would result from a comprehensive analysis considering all private and public costs and benefits. We used capital costs for energy equipment and renovations based on mean values from the NREL National Residential Efficiency Measures Database (NREMD) 67 . In some cases, capital costs for the precise technology deployed in a renovation measure were not available in this database and for such cases we defined proxy capital costs, on the basis of the ranges of costs that are present in the database. A full list of the assumed costs and indication where costs were assumed due to missing values in the database is available on the archived code repository 68 . It is important to note that renovation costs can vary widely case by case and our use of average values does not incorporate that variation 69 . As renovations are implemented on the basis of observed empirical renovation rates for different renovation types, it is assumed that each piece of energy equipment is replaced at end-of-life. Thus, the capital cost used for the NPV calculation is the difference in cost between the new equipment type (for example, a heat pump) and a replacement of the same equipment type being replaced (for example, a gas furnace). Heat pump replacements are assumed to replace heating equipment only, not combined cooling and heating equipment, which would improve the NPV of such renovations. Envelope renovations are priced by the material and installation costs per area of the building that receives a certain type of insulation (for example, external wall, basement ceiling and roof). The cost of an envelope renovation is calculated as the cost of going to the postrenovation state (for example, R-15 wall insulation) minus the cost of installing the prerenovation state (for example R-7 wall insulation). Thus, costs for homes with little/no prerenovation insulation will be higher. We assume no difference in renovation costs by building age. Future retail energy prices (in US$ 2021) by Census Division were taken from the reference case of EIA's annual energy outlook 70 . Our cost-benefit analysis is restricted to renovations taking place by 2025 for two reasons: first, fuel price projections are not available past 2050 (meaning a cost-benefit analysis with a 25-year horizon cannot be calculated for investments later than 2025). Second, future energy equipment costs are uncertain and may change considerably from the values in the NREMD database. For instance, heat pump unit costs may decline with large increases in sales 71 .
The NPV of energy renovations was estimated using a 3% real discount rate. The analysis incorporates future electricity GHG intensities at the GEA region level. Results of GHG reduction potential, NPV and GHG abatement costs (calculated as −1 multiplied by the NPV divided by the reduction in GHG emissions over the equipment lifetime) are calculated and shown in Supplementary Section 3.6, using the projected GHG intensities from the LREC scenario only.

Limitations.
Here, we draw attention to several limitations of our modelling approach. Similar to any prospective scenario analysis, there are uncertainties inherent to the model input parameters, which grow larger as the model gets further into the future. In place of sensitivity analyses to assess the uncertainty around each input parameter, we generated a large scenario space by selecting feasible ranges of input values for parameters considered to be influential on Extended Data Fig. 2 | GHG emissions reduction by sequential strategy adoption. Mitigation actions beyond the Baseline scenarios (black dashed line) are grouped into strategies affecting electricity supply (blue), renovation of existing homes (orange), and housing stock evolution (HSE)/new housing characteristics (NHC) (pink/purple). a) Strategies groups are ordered according to greatest cumulative emission mitigation potential. b) Strategy groups are ordered according to reverse cumulative emission mitigation potential. Fig. 4 | CO 2 intensity of electricity generation, 2020-2050. Cambium Generation and Emission Assessment (GEA) regions 55 are grouped in two groups based on alphabetical order, to facilitate legible legends. a) Group 1 Mid-Case electricity supply, b) Group 1 Low Renewable Electricity Cost electricity supply, c) Group 2 Mid-Case electricity supply, d) Group 2 Low Renewable Electricity Cost electricity supply. Further reductions of CO2 intensity of electricity were assumed beyond 2050, as described in Supplementary Information Section 5.