Global greenhouse gas emissions from residential and commercial building materials and mitigation strategies to 2060

Building stock growth around the world drives extensive material consumption and environmental impacts. Future impacts will be dependent on the level and rate of socioeconomic development, along with material use and supply strategies. Here we evaluate material-related greenhouse gas (GHG) emissions for residential and commercial buildings along with their reduction potentials in 26 global regions by 2060. For a middle-of-the-road baseline scenario, building material-related emissions see an increase of 3.5 to 4.6 Gt CO2eq yr-1 between 2020–2060. Low- and lower-middle-income regions see rapid emission increase from 750 Mt (22% globally) in 2020 and 2.4 Gt (51%) in 2060, while higher-income regions shrink in both absolute and relative terms. Implementing several material efficiency strategies together in a High Efficiency (HE) scenario could almost half the baseline emissions. Yet, even in this scenario, the building material sector would require double its current proportional share of emissions to meet a 1.5 °C-compatible target.


Model framework
The overall model framework is shown in Supplementary Figure 1. It is comprised of a building demand module, a material demand module, and a material supply module. Each module is described in turn.
Supplementary Figure 1. The conceptual framework of the GloBUME model. This dynamic GloBUME model (global dynamic building materials use and embodied GHG emissions model) consists of three layers, i.e. a building demand layer, a material demand layer, and a material supply layer. Two rectangles in the center represent stocks of floor area and materials respectively, ovals denote flows, hexagons are the drivers or determinants. Colors indicate drivers given by each scenario, driven by related strategies illustrated in the rounded rectangles filled with the same colors. Data and codes details are freely available from the link: https://doi.org/10.5281/zenodo.5171943.

Building demand layer
In the building demand layer, the dynamic building stock and flows are calculated by the following equations: As shown in equation (3), ′ ( ′ < ) represents the year in which the building was constructed, and ( , ′ ) denotes the probability that the buildings built in year ′ will be demolished in year . We assume buildings' lifetime follows the Weibull distribution and use the equation (4) to calculate ( , ′), where ( ) and ( ) represent the scale parameter and the shape parameter in year , respectively.
Note that the functional form for the lifetime distributions can have a considerable impact on the calculation of outflow and inflow 1 . Other functional forms, e.g., normal 2 , Log-normal 3 , and Gamma functions 4 have also used in material related dynamic analysis. However, the Weibull distribution is the most common in the literature, and is therefore the distribution for which the most reliable data are available 5 .
This study covers 4 types of residential buildings (detached houses, semi-detached houses, apartments, Note that although we mainly consider social-economic factors (i.e., population and floor area per person) as the driver of the global demand in building stocks. There are other factors that will impact future building stocks and could be modelled in future work (using different modelling approaches). For example, the increasing frequency and severity of climate changes may create extra building demand as a complement for those damaged by natural disasters 8 , or abandoned in areas declared uninhabitable 9 .

Material demand layer
Materials inflow in year , defined as ( ) , is calculated by multiplying building inflow with materials use intensity per floor area in year , denoted as ( ). We include seven major building materials, i.e. steel, concrete, brick, aluminium, copper, wood, and glass. Detailed material density data is included in the Scenario generation section (see below).
As shown in equation (6), the materials outflow in year , denoted by ( ), is determined by the cumulative sum of the floor area of demolished buildings in year (i.e. ( , ′ ) * ( ′ ) ), and the material intensity in year ′ (i.e. ( ′ )), when the demolished buildings were built.

Material supply layer
Materials demand can be met by primary production, secondary production, or reuse and their emission intensities differ significantly (cradle-to-gate emission intensity is 0 for reused materials). In equation (7), ( ) and ( ) denote the quantities of materials supplied by primary and secondary production respectively, while ( ) and ( ) represent the GHG emission intensity of the cradleto-gate life cycle primary and secondary material production, respectively. ( ) and ( ) can be calculated by equation (8) and (9) respectively, where ( ) and ( ) represent the reuse and 6 recycling rate of building materials, i.e., the percentage of post-consumer material scraps that are collected for reuse or to produce secondary materials, respectively. Note that here are some inherent assumptions: 1) we only consider the recycling and reuse of the newly generated material outflows, which means the historical scraps would not be considered for recycling and reuse in the future; 2) we only consider the recycling and reuse of material outflows within the same region, area, and building type, which means cross-regional scraps, for example, are not considered for recycling and reuse; 3) we assume that the secondary materials provide the same service as the primary materials, that is, a 1:1 substitution between primary and secondary materials in mass terms. Under these assumptions, our circularity model should be taken as a simplified approach for short-term scenario analysis where material inflows are generally larger than outflows. However, in the longer-term (i.e. beyond 2060) when material outflows largely outstrip inflows, it could be important to consider the potential of using historical scrap reserves and scraps moving across regions/areas/building types.

Scenario generation
Our model described above provides an approach of scenario explorations based on customized development trends. Note that our investigations here represent what-if scenarios rather than an accurate prediction of the future. We follow Deetman et al., 2020 5 , in assuming that some variables (e.g., building lifespan, and material intensity) have remained constant in the pre-2020 period based on 2020 levels (for more details see Deetman et al., 2020 5 ). For variables where historical data is available (e.g., for population, and GDP) year-specific data was used. For the future, the Baseline scenario represents a reference case where the historical trends in the building sectors around the world largely continue. The High Efficiency scenario represents a deep emission mitigation scenario where the seven strategies will be simultaneously adopted. Please see below for the full details of the strategies in each scenario.

M1-More intensive use
More intensive use represents the potential to decouple the housing demand (per capita floor area) from economic development. In the Baseline scenario, the residential floor area per capita sees increasing trends, in line with the IMAGE-SSP2 baseline scenario 6 (IMAGE data is allocated to the four building  5 and then extended to 2060 using the same regression model as that used in the original study 5 . In the High Efficiency scenario, we adopt the assumption from the literature that sees a potential for 20% reduction 12 from the Baseline values by 2050. It is noteworthy that in the High Efficiency scenario, in general, the floor area per capita will still increase (at a lower speed than in the Baseline scenario) in the rapidly urbanizing regions but slightly decrease in the highly urbanized regions.
As such, the floor space gap between the rich and poorer regions will be narrowed. In future work, it may be worth exploring a more realistic and region-specific implementation of a more-intensive-use scenario, for example by adjusting the per capita floorspace differently for urban and rural areas. This would ideally account for the various drivers of floorspace demand trends (such as high demand & prices in densely populated areas) and make sure that assumptions are not at odds with achieving decent living standards or the Sustainable Development Goals 13 .

M2-Lifetime extension
Short-lived buildings lead to frequent demolition of old buildings and construction of new buildings. The construction activities are associated with large amounts of building materials and GHG emissions.
Therefore, prolonging service life of buildings can reduce the construction of new buildings and therefore avoid materials consumption and emissions 14,15 . Today, the service life of buildings is fairly short and sees a potential of up to 90% extension by 2050 12  Under the Baseline scenario, we assume that the lifetime of buildings will stay at the current level. In the High Efficiency scenario, we assume a lifetime extension by 30%, 60%, and 90% by 2050 for residential buildings newly constructed in the regions classified as having long-, medium-, and short-lifetimes, respectively (see Supplementary Table 4). Similar assumptions are found in the literature 12 , which assumed a 90% lifetime extension by 2050 in several countries. We assume a lifetime extension by 60% for newly constructed commercial buildings by 2050. Supplementary

M3-Lightweight design
Today's construction practices often overuse materials due to less efficient design or a relatively low physical strength of primary materials 17,18 . There is a potential for lightweighting through more advanced design strategies including novel structural design 19 , typology optimization 20 , additive construction (such as 3D printing) 21 , and high strength steel and aluminium utilization 22 . For the Baseline scenario, we assume that material use intensity will stay at the current level. We used the material density of steel, Note: *In practice the regional difference in brick intensity can be considerable. The model can be improved when more data are available.

M4-Material substitution
Timber houses are generally considered an environmentally friendly substitution for steel and concrete structures for two reasons: firstly, using timber can avoid considerable emissions during material production processes 39 ; secondly, timber can serve as long-term carbon storage if sources sustainably 40 .
However, there are also concerns surrounding land-use change to manage forests and related biodiversity loss 41 48 . Based on these estimates, we adopt 20% in (and prior to) 1950 to 89% in 2020, following a linear increase.
In the Baseline scenario, we assume that the adopted recycling rate of steel, copper, and aluminium in 2020 will be constant until 2060. Evidence shows that the recycling rate of steel, copper, and aluminium has a potential to reach 90% 49 , 93% 45 , and 95% 48 respectively. In the High Efficiency scenario, we assume a linear increase from 2020 values to the potential maximum recycling rates from the literature by 2050 and they are held constant afterwards. See a summary in TableS.7.
There is a potential for increasing the reuse of building components through increased prefabrication and modular construction design 12 . Based on case studies, we assume that up to 15% 18 of steel and concrete could be reused with these approaches. Both assumptions are somewhat conservative compared to the recent scenario analysis for several countries 12 , yet were chosen considering the varying technology and regulation development patterns across global regions. In the Baseline scenario, no reuse is adopted for any material. In the High efficiency scenario, we assume a linear increase to 15% reuse for concrete and steel between 2020 and 2050 and hold constant afterwards. Note that an inherent assumption is that reused components will meet the function needed during the service life of the building. While this may be quite reasonable in our scenarios up to 2060, it may not be so for later in the century. Split lifespans of the buildings and reused components may be considered when exploring scenarios of longer time horizon, e.g., towards and beyond 2100.

M6-Energy transition
The trend in decarbonization of the global energy system may have significant impacts on the GHG emissions of materials production 50 . Among all energy types, the electricity system is expected to see profound changes in the coming decades 50 52 . These electricity scenarios have incorporated consistent techno-socio-economic development in both energy demand (e.g., population size, income-level, and lifestyle) and energy supply (e.g., costs of competing electricity generating technologies) determinants 51 . Specifically, we apply the electricity scenarios that are compatible with SSP2-baseline and SSP2-2.6 pathways to our Baseline and High Efficiency scenarios, respectively. The SSP2-2.6 corresponds to a strong mitigation scenario aligned to the RCP2.6 (Representative Concentration Pathway), and thus realizes a radiative forcing of 2.6 W/m2 in 2100 (approximately 450 ppm CO2eq.) 53 . Biogenic carbon flows are not included in the IPCC GWP method used, which means that no credit is given to BECCS. As such, our evaluation of the emission reduction from the electricity transition may represent a lower bound. Also note that we do not take into account the transition in heat and other fuel types due to a lack of data. 2020 is set as the base year for all electricity transition scenarios. That is to say the electricity system changes apply only after 2020.

M7-Production routes and efficiency improvement
The literature indicates considerable potentials for improving the efficiency of material production processes or switching to different material production routes in the future 50,54 . Such developments can reduce energy requirements of future material supply and thus also associated emissions 50,54 .
For the Baseline scenario, we assume the production efficiency and the production routes will not change in the future. For the High Efficiency scenario, we consider scenarios for two developments in this study.
These are energy efficiency improvements and a change in market share of the primary copper production routes. First, energy efficiency increases are modeled for the production of steel, aluminium, and copper.
For steel making, an improvement of 1.5% per year is assumed based on the study by Van

Ore grade decline
There has been a secular ore grade declined over the last century for copper 59 , nickel 60 , zinc 61 , and lead 61 due to large amounts of metal mining across the globe. A lower ore grade means that more ore needs to be processed to produce the same amount of metals 59 . This leads to increased energy requirements and higher emission intensities of primary metal supply 50 . Since copper, nickel, zinc, and lead are amongst the top seven metals consumed globally 62 , their impacts play a considerable role in emission intensities of numerous production processes in the downstream supply chain 45 . To include their impacts on the building material production in the future, we consider ore grade decline scenarios and incorporate them equally into the Baseline and High Efficiency scenarios.
We derive metal-specific scenarios of future ore grades between 2020 and 2050 from Harpprecht et al.

Evaluation of materials demand by scenario
Using Python, we develop a cohort-based and stock-driven dynamic building materials model to evaluate materials demand in the building sector. On this basis we assess the effect of four strategies (M1-M4, see above) on building materials demand towards 2060. A detailed introduction of the dynamic model can be found in the Model framework section above and the Methods section in the main text.

GHG emission intensity of building materials by scenario
We evaluate the GHG emission intensity of each building material using a prospective LCA model as described below. The scope is the crade-to-gate production system of per kg of each building material.
We use the ecoinvent 3.6 cut-off database 64 as the life cycle inventory database because of its global coverage and regional resolution. The ecoinvent processes used are shown in Supplementary Table 9. In general, the regional mapping in ecoinvent has a different resolution with that in IMAGE. According to the approach used in Mendoza Beltran et al. (2020) 51 , for ecoinvent regions that involve more than one region in IMAGE, we apply the LCA results for the larger region to all smaller regions included. For IMAGE regions involving more than one ecoinvent region, we use an average of the ecoinvent data. As an example, Supplementary Table 10 shows the regional match for the primary copper case.
For the impact category of GHG emission, we use the Global Warming Potential per substance from the IPCC 2013, with a time horizon of 100 years 65 . We use the Activity Browser (AB) software 66 to conduct the prospective LCA calculation because of its high-efficiency at scenario-based modeling. To model the development of GHG emission intensity of different unit processes in the different scenarios, we need to incorporate the relevant scenario data into the LCI database of ecoinvent. For the Baseline scenario, we incorporate the ore grade decline data into the ecoinvent database. For the High Efficiency scenario, data on ore grade decline, energy transition, efficiency improvement, and production routes development are incorporated into the ecoinvent database. To do this, we translate the scenario data of each parameter into the presamples structure 67 required for the prospective LCA calculation in the Activity Browser 66 . In practice this means that the scenario data of each parameter is specified in excel files using the approach developed by Steubing et al. (2020) 66 . We also conduct a 'Contribution analysis' using the Activity Browser to split the contribution of CO2 and non-CO2 emissions to the total GHG emission.
Supplementary in RER (Europe) and RAS (Asia) from ecoinvent (for primary copper production).

Additional figures and tables
Supplementary Figure 3. Building material outflow by magnitude during 1980-2020 (left axis) and recycling input rate (the ratio of available recycled metals from outflows to annual inflows, for steel, copper and aluminium in dashed line) (right axis). Supplementary

Uncertainty and sensitivity analysis
The aim of this research is to evaluate long-term materials and emission trends across different regions and income groups. We then investigate the emission mitigation potentials for key material efficiency strategies. We are not able to explicitly evaluate the uncertainties across all datasets due to data limitations. However, we conduct two analyses to understand where major uncertainties may arise so as to facilitate improvements for future work.
First, we compare several primary variables in the Baseline scenario with other literature (Supplementary Table 12). For example, we employ the population trend of IMAGE-SSP2 as a representative of a middleof-the-road path, which is comparable to the recent scenario explored by Vollset et al., (2020) 70  Second, we employ a sensitivity analysis to show how the High efficiency scenario is influenced by different assumptions on several variables: more intensive use, lightweight design, lifetime extension, material substitution, and more recovery. We assume a sensitivity range for each variable (see Supplementary Table 12). Note that for light weight design, increased recycling and reuse, the High strategies on GHG emissions. The green lines in all panels represent the High efficiency scenario.
More details are available in Supplementary Table 13. Supplementary