Housing stock evolution and new housing characteristics scenarios are based on scenarios developed by Berrill and Hertwich22, which is the source of estimated embodied emissions from material production and construction, and where a full description of the housing stock model (HSM) can be found. County population projections54 drive the housing stock model22, and are scaled to the mid-range scenario from the most recent U.S. Census Bureau population projections to 206055. The scenarios are extended for this work to include the new housing characteristics scenario of increased electrification in new housing, and to describe scenarios renovation of existing housing. The HSM is run at the resolution of US counties, and incorporates dynamics of region- and house-type-specific vacancy rates, which influences local demand for new construction. Vacancy rates by Census Region and house type are assumed to gradually converge, which translates into higher construction rates in counties with comparatively low vacancy rates. The Baseline stock scenario (1) assumes a continuation of historic loss rates, which are defined for each house type, age-range, Census Division, and vacancy status combination. High stock turnover (scenario 2) is simulated by increasing housing stock turnover rates by a factor of 1.5. High multifamily stock growth (scenario 3) is a scenario representing both sufficiency (as it lowers growth in floor-area per person) and more intensive urbanization, facilitated in part by lowering regulatory barriers to multifamily construction in urban centers. This is simulated by increasing the county multifamily population share by 0.25 percentage points annually in counties with population growth of at least 5% over twenty years, for two periods, 2020–2040, and 2040–2060. For instance, a county with a multifamily population share of 20% in 2020 and sufficiently high population growth to 2060 will see their multifamily population share grow to 30% by 2060. This approach avoids increasing of multifamily population share in counties with low or negative population growth, which we consider to be less likely.
Emissions from material production and onsite energy and transport in new construction are calculated for 51 housing archetypes22, capitalizing on high resolution representation of US housing characteristics by house type, size, foundation type, heights, etc., in the ResStock housing characteristics data46. 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 construction22. Embodied emissions from energy equipment such as furnaces and heat pumps are not considered. It is worth emphasizing that increased emissions from higher stock turnover are estimated at the entire stock level. It is probable that for certain homes with large renovation challenges which can be replaced with relatively low embodied emissions, demolition and rebuilding may be a more practical solution to lower life-cycle emissions. The baseline housing stock growth scenario already projects increased construction in many urban areas, due to an assumption of converging vacancy rates by house type and Census Region22. Higher housing stock growth in areas with low vacancy rates, which are largely urban areas with stringent land-use restrictions47, could help to alleviate issues of housing affordability and supply56, which are outside of the scope of the current analysis.
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 five-year periods spanning 2021-2060 (2021-2025, 2026-2030, etc.). Future housing characteristics are modified depending on anticipated adoption of residential building energy codes by states57, updates to federal energy appliance standards58, and assumptions on electrification and efficiency improvement of equipment and insulation. Building energy codes mostly apply to building envelope characteristics, such as insulation and infiltration levels, energy ratings of windows, etc.59, while the federal efficiency standards apply to energy consuming equipment and appliances such as space and water heaters, air-conditioning systems, refrigerators, etc. We also incorporate assumptions regarding changes and trends in housing and energy appliance characteristics that are not directly based on codes and standards, but more related to household preferences and energy and appliance prices, 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 Base 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. Increases in electrification of new housing is defined for four Census Regions, based on projected price differences between electricity and natural gas60. Price differences are largest (and electrification rates assumed to be lowest) in the Northeast, while price differences are smallest (and electrification rates assumed to be highest) in the South. In between are the West and Midwest Regions, which are assumed to have similar moderate electrification rates, with West slightly higher than Midwest in the 2020s. While the rates of change are defined by Census Regions, individual characteristics are specified, and change, at higher spatial resolution. For instance, heating fuel characteristics are defined for every PUMA61, so PUMAs with initially low shares of natural gas (for instance due to absence of gas distribution networks) will continue to have low shares of natural gas in new construction, which decline over time in line with the scenario.
In the Reduced Floor Area scenario (B), no new housing unit exceeds a size of 279 m2 (3,000 ft2), an arbitrary limit which is chosen based on the floor area bins used in the ResStock housing characteristics database. This scenario represents a sufficiency measure aimed at limiting growth of floor area per person, by replacing ‘very large’ new homes with moderately large new homes (Supp. Fig. 17). Housing that previously fit into the two largest size categories of 279-371 m2 (3,000–3,999 ft2) or 372+ m2 (4,000+ ft2) are reassigned to be in one of the 186-232 m2 (2,000–2,499 ft2) or 232-279 m2 (2,500–2,499 ft2) ranges with 50:50 probability (Supp. Fig. 21). Multifamily and manufactured housing are essentially unaffected by this change, as very few of those housing types exceed 279 m2, but for single-family housing, this scenario reduces mean floor area of new houses by 25%, from 258 m2/house to 193 m2/house22 (Supp. Fig. 18). One alternative approach to modelling scenarios of reduced floor area in new housing would be to increase the share of homes built in the smallest size ranges, below 93 m2 (1,000 ft2), representing growth in accessory dwelling units currently seen in urban areas of the US with housing shortages62. In the Increased Electrification scenario (C), electrification of new housing is much more rapid, with all Regions reaching complete electrification by 2030 except the Northeast, which is fully electric by 2040 (Supp. Fig. 16). The Increased Electrification and Reduced Floor Area scenario (D) simply combine the new housing characteristics scenarios B and C. Further information on new housing characteristics scenarios is provided in section 3 of the Supplementary Information.
Our analysis represents the most comprehensive 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 (PUMA) level and empirical data on recent renovation trends, and estimating energy reductions of renovation actions with a detailed physical simulation model. 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 renovation categories capture the main types of retrofits which offer substantial potential for energy reductions46. Two pieces of information are required for each renovation, the rate of renovation in the housing stock (i.e. the probability of a housing unit making a specific type of renovation in a given year), and the characteristics of a given system post-renovation, conditional on its pre-renovation status. Our estimates of future renovation rates and fuel-switching trends are based on data from American Housing Surveys (AHS) covering the period 1995-201963, which include information on whether homes replaced or added central AC, space heating equipment, water heaters, or insulation. These questions were only asked of owner-occupied households. Without specific data for tenant-occupied households, we assume that the renovation rates and characteristics identified for owner-occupied homes apply to all homes.
We define three renovation scenarios, which we term ‘regular’, ‘advanced’, and ‘extensive’. The standard renovation 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 advanced renovation 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 higher shift towards electric space and water heating systems, and heat pumps in particular. In the extensive renovation scenario, much higher rates of electrification of space and water heating takes place, with 100% of replacements of fossil heating renovations with electric heat pumps from 2025 on. This does not mean that all fossil heating equipment is replaced by heat pumps in 2025, but if a fossil-based 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 Information Section 2. Renovations are only applied to housing built before 2020, the energy efficiency of homes built between 2020-2060 is assumed to remain static (Supp. Table 6). Apart from the effect of replacing fossil heating equipment with heat pumps, neglecting renovation of newly built homes is not expected to have an influence on overall results, as renovation of new homes would only start to take effect from ~2040 onwards, and most homes built from 2020 onwards already have relatively high efficiency.
We calculate energy related GHG emissions using standard emission factors for combustion of fossil fuels64, and annual average CO2 intensity for three electricity supply scenarios; Mid-Case (MC) and Low Renewable Energy Cost (LREC) from NREL’s standard scenarios24, and a scenario involving 100% Carbon Free Electricity (CFE) by 203525 (Supplementary Figure 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 a detailed electricity supply scenario detailing electricity generation by source, we assume the same intensities as LREC until 2025, which then half between 2025 and 2030, before reaching zero by 2035. Electricity GHG intensities are combined with electricity consumption calculations at the level of 18 regional transmission organizations (RTOs). Regarding coverage of different GHGs from different stages of supply chains, energy-related emission intensities describe CO2 emissions from combustion only24, 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-CO2 combustion products, but CO2 emissions account for over 99% of total combustion GHGs64. Embodied emissions from material production and construction22 are based on material life cycle assessment databases, environmental product declarations and literature, and include non-CO2 GHGs. To estimate the land requirements of wind and solar generation in the electricity supply scenarios24, we divide the generating capacity of onshore wind, utility PV, and distributed PV65 by technology-average power density coefficients for renewable electricity in the US37; 3.1 We/m2 for onshore wind, 5.8 We/m2 for utility PV, 6.7 We/m2 for distributed PV, and 9.7 We/m2 for concentrate solar power. Our estimates of land area for wind generation capacity do not include the offshore area required for offshore wind.
Calculation of energy consumption in the US housing stock is done 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 question46,61. 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. Without a reliable approach to estimate, or reduce energy consumption in vacant housing, we do not incorporate 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 based on energy consumption by energy carrier in each year, and are interpolated for the intervening years in which energy demand is not simulated (e.g. 2021-2024) using the spline() function in R. In order to capture the heterogenous characteristics of the US housing stock in a representative manner46, we simulate energy consumption in a large number of houses for each scenario and simulation year, so that one simulation represents somewhere in the range of 590-800 homes. 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 outputs22.
Supplementary 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 occur66. As a basis for the Paris 2050 target and 50% 2030 target in Figure 1, we calculate total residential emissions in 2005 by combining residential energy emissions67 with emissions from investment in new housing in 2005, using data from Berrill et al.11, scaled by 0.99 to exclude Alaska and Hawaii. As a basis for the 1.5°C target in 2030, we use total 2020 residential emissions as calculated by our model.
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 usually grow larger as the time period gets further into the future. In place of sensitivity analyses to assess the uncertainty around each input parameter, for this study we generated a large scenario space by selecting ranges of input values for parameters considered to be influential on future emission trajectories, such as the rate and depth of renovations, decarbonization of electricity supply, etc. 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 however intended to represent all possible future emission pathways. Parameter values excluded from our scenarios space which would likely result in notable differences to the range of emission pathways estimated include higher or lower population and housing stock growth trajectories, electricity supply scenarios resulting in slower decarbonization of electricity, and slower renovation rates. A rather pessimistic scenario, assuming fixed electricity GHG intensity at 2020 levels and no renovation of existing housing, is however included in our illustration of annual emissions 2020-2060 in Figure 4. This could reasonably be considered as a worst-case outcome for future emissions, and shows almost no change in the level of annual emissions over the next forty years.
Our calculations of embodied emissions from construction are based on a previous study22, and incorporate moderately optimistic assumptions on reduction in GHG-intensity of material production, which reduces emissions per m2 floorspace by on average 23% from 2020-2060. More ambitious 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 construction29, 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 even with much faster decarbonization of construction.
To keep the number of required simulations to a reasonable level, we only assess renovations to housing built until 2020, none of the new housing built 2021-2060 are assumed to undergo renovation. This results in overestimations of energy emissions from new housing, particularly in the later decades of the analysis. However, as average energy intensity is already lower in housing built after 2020 than existing housing, pre- and post-renovation (Supplementary Tables 5-6), and as even in 2060 most energy-related emissions come from existing housing (Figure 3), this simplifying assumption is considered to have negligible influence on our results and overall findings. Scenarios with increased electrification of new housing can serve as a proxy for futures in which the housing stock built after 2020 undergoes greater electrification-focused renovation.
Finally, in the estimation of GHG emissions from electricity generation we use projections of annual average emissions for 18 regional transmission organizations (RTOs). Although Standard Scenario GHG intensities are available at a spatial higher resolution of 134 local balancing areas39, the larger RTOs were preferred to represent the average energy supply mix of electricity consumed by households. Further, annual average emission intensities were used instead of short-run or long-run marginal emission rates, as the annual average were readily available from the Cambium scenario data downloader65. Using marginal emission rates may be more suitable when considered the time of day of residential electricity demand vis-à-vis electricity demand from other sectors, or when considering the longer term growth in electricity demand from residential buildings compared to other sectors. Such temporal considerations and intersectoral interactions were outside of the scope of the present analysis, and represent a promising avenue for future research considering increased electrification in all sectors39.