Modeling Material Stocks and Embodied Emissions from the Built Environment: A machine learning and dynamic material stock and flow framework
Doctoral thesis, 2026
The building and construction sector is responsible for around 37% of global carbon emissions. It is therefore crucial to dicarbon the sector, and it is currently not on track to meet the Paris Agreement aligned decarbonization pathways. The carbon emissions from building and construction can be classified into operational emissions from direct energy consumption, and the embodied emissions associated with the material production process. Policy efforts have been mainly focused on reducing operational emissions through measures to reduce energy consumption by building renovation such as the Energy Performance of Buildings Directive (EPBD) in the European Union. However, prioritizing renovation could result in trade-offs such as increased embodied emissions.
It is therefore important to understand and model the embodied emissions from the building and construction sector, especially regarding maintenance and renovation. A key challenge in developing detailed bottom-up material stock and flow model for the built environment is the lack of data. This thesis addresses this challenge by developing a modeling framework that applies machine learning to predict and impute missing inventory data, and subsequently further develops material flow analysis models to better estimate embodied emissions from maintenance and renovation. This modeling framework is first developed using roads in Sweden as a case study and further developed and refined for residential buildings in Sweden. The machine learning approach is developed to circumvent the need for scarce height data to predict usable floor space and construction year for buildings. The building inventory dataset is then complemented by a detailed material intensity (MI) dataset that can be disaggregated into building layers, and a material flow model that ensures renovation is correlated with demolition is developed. Lastly, the modeling framework is integrated with a building energy model to analyze the total emissions reduction potential for energy efficiency renovation for the Swedish city of Uddevalla.
The results show that firstly machine learning can be effectively applied to predict inventory data in absence of height data, and such approaches should be tested in other geographical areas. The embodied emissions results show that in Sweden embodied emissions from maintaining roads and renovating residential buildings (both functional and energy efficiency renovations) is expected to exceed the embodied emissions from new construction. Therefore, more policy attention is needed to reduce embodied emissions from renovations.
It is therefore important to understand and model the embodied emissions from the building and construction sector, especially regarding maintenance and renovation. A key challenge in developing detailed bottom-up material stock and flow model for the built environment is the lack of data. This thesis addresses this challenge by developing a modeling framework that applies machine learning to predict and impute missing inventory data, and subsequently further develops material flow analysis models to better estimate embodied emissions from maintenance and renovation. This modeling framework is first developed using roads in Sweden as a case study and further developed and refined for residential buildings in Sweden. The machine learning approach is developed to circumvent the need for scarce height data to predict usable floor space and construction year for buildings. The building inventory dataset is then complemented by a detailed material intensity (MI) dataset that can be disaggregated into building layers, and a material flow model that ensures renovation is correlated with demolition is developed. Lastly, the modeling framework is integrated with a building energy model to analyze the total emissions reduction potential for energy efficiency renovation for the Swedish city of Uddevalla.
The results show that firstly machine learning can be effectively applied to predict inventory data in absence of height data, and such approaches should be tested in other geographical areas. The embodied emissions results show that in Sweden embodied emissions from maintaining roads and renovating residential buildings (both functional and energy efficiency renovations) is expected to exceed the embodied emissions from new construction. Therefore, more policy attention is needed to reduce embodied emissions from renovations.
Material flow analysis
Embodied emissions
Material stock
Machine learning
Author
Qiyu Liu
Chalmers, Space, Earth and Environment, Energy Technology
Development of a machine learning model to improve estimates of material stock and embodied emissions of roads
Cleaner Environmental Systems,;Vol. 14(2024)
Journal article
Predicting building age and floor space using feature-engineered 2D urban morphology
Smart and Sustainable Built Environment,;Vol. In Press(2026)
Journal article
Liu Q, Lanau M, Rootzén J, Cao Z, Johnsson F. A Layered Dynamic Material Flow Framework for Modeling Building Renovations.
Liu Q, Lanau M, Somanath S, Rootzén J, Johnsson F. Dynamic assessment of the climate benefits of biobased insulation materials.
The Swedish building and construction sector is responsible for a large amount of greenhouse gas emissions, and the sector needs to be urgently decarbonized for Sweden to become carbon neutral by 2045. Emissions from the sector can be grouped into two main types: operational emissions from direct energy use such as heating a building and embodied emissions from the production of construction materials such as steel and cement. In Sweden, embodied emissions are the bigger challenge to decarbonize as the Swedish electricity and heat supply mix are relatively low in the use of fossil fuels.
To better develop science-based policy to reduce emissions from the building and construction sector, it is important to be able to model and estimate when these emissions will occur and how much emissions are expected to occur. The main stages when embodied emissions are generated in a road or building’s life cycle are the new construction stage and the renovation/maintenance stage. The emissions from the renovation/maintenance stage are especially important to understand as most current policies target the new construction stage. In addition, policies that target renovations such as the EU’s Energy Performance of Buildings Directive (EPBD) do not include embodied emissions from renovations. Therefore, there is a need to develop a modeling framework to estimate the embodied emissions from renovations.
The established method to estimate such embodied emissions is dynamic material stock and flow analysis. The model framework first accounts for how much material is accumulated in a stock (e.g., buildings) and then estimates the material flow from renovation or maintenance by applying a statistical likelihood of when the stock will reach its end of life. The key challenge to conducting detailed material stock and flow analysis is the lack of data.
This thesis develops a modeling framework that first utilizes machine learning to predict missing data necessary to calculate material stock and then further develops existing material flow methods to model renovations in a greater detail. The framework is first applied to road maintenance in Sweden and then further developed in more details for residential buildings in Sweden. The results show that machine learning can be effectively utilized for predicting data for material stock modeling. In terms of emissions, the modeling demonstrates that embodied emissions from residential renovations exceeds the embodied emissions from new construction of buildings and thus more policy attention should be paid to reduce embodied emissions from renovations in Sweden.
To better develop science-based policy to reduce emissions from the building and construction sector, it is important to be able to model and estimate when these emissions will occur and how much emissions are expected to occur. The main stages when embodied emissions are generated in a road or building’s life cycle are the new construction stage and the renovation/maintenance stage. The emissions from the renovation/maintenance stage are especially important to understand as most current policies target the new construction stage. In addition, policies that target renovations such as the EU’s Energy Performance of Buildings Directive (EPBD) do not include embodied emissions from renovations. Therefore, there is a need to develop a modeling framework to estimate the embodied emissions from renovations.
The established method to estimate such embodied emissions is dynamic material stock and flow analysis. The model framework first accounts for how much material is accumulated in a stock (e.g., buildings) and then estimates the material flow from renovation or maintenance by applying a statistical likelihood of when the stock will reach its end of life. The key challenge to conducting detailed material stock and flow analysis is the lack of data.
This thesis develops a modeling framework that first utilizes machine learning to predict missing data necessary to calculate material stock and then further develops existing material flow methods to model renovations in a greater detail. The framework is first applied to road maintenance in Sweden and then further developed in more details for residential buildings in Sweden. The results show that machine learning can be effectively utilized for predicting data for material stock modeling. In terms of emissions, the modeling demonstrates that embodied emissions from residential renovations exceeds the embodied emissions from new construction of buildings and thus more policy attention should be paid to reduce embodied emissions from renovations in Sweden.
MISTRA Carbon Exit Phase 2
The Swedish Foundation for Strategic Environmental Research (Mistra) (MISTRACarbonExitPhase2), 2021-07-01 -- 2025-03-31.
The Swedish Foundation for Strategic Environmental Research (Mistra) (2016/11), 2025-04-01 -- 2026-09-30.
Driving Forces
Sustainable development
Subject Categories (SSIF 2025)
Construction Management
Energy Systems
DOI
10.63959/chalmers.dt/5926
Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 978-91-8103-469-1
Publisher
Chalmers