Method development to improve estimates of embodied carbon emissions in the Swedish built environment

Licentiatavhandling, 2025

The building and construction sector accounts for approximately 20 percent of Sweden’s total greenhouse gas (GHG) emissions. To achieve Sweden’s climate goal of reaching net-zero greenhouse gas emissions by 2045, urgent action is thus needed to decarbonize the building and construction sector. Emissions from the building sector can be classified into two sources: 1) operational emissions related to energy consumption, and 2) embodied emissions from the manufacturing and processing of building materials. The share of embodied emissions as a percentage of the total emissions is expected to increase as more efforts are put into energy efficiency measures and decarbonization of the energy supply. As a result, more attention is needed to reduce embodied emissions from the building and construction sector.

However, embodied emissions in the built environment remain relatively understudied, especially at the national level. Embodied emissions from the built environment are most commonly studied using material stock and flow analysis. Material stock and flow analysis can be classified into top-down and bottom-up, and this study focuses on the latter. Top-down analysis typically relies on aggregated national or sectoral data, such as economic or material flow accounts, while bottom-up analysis is based on detailed data at the building or component level. Existing bottom-up studies primarily focus on small geographical scales, such as neighborhood and city levels, which limits the applicability of their findings national level policy making. The primary challenge in conducting national-level estimations of embodied emissions lies in the limited availability of inventory data. Moreover, maintenance and renovation activities are frequently overlooked in models of material flows within the built environment.

This thesis addresses this gap by developing a framework to estimate embodied emissions from the built environment at the national level while having limited available data. The challenge of lack of available data is tackled by using machine learning (ML) models. Paper I estimates the material stock, flow, and embodied carbon from Swedish roads while predicting missing road widths data using a ML regression model. Paper II expands the scope to residential buildings by predicting construction years of buildings using a classification ML model and usable floor space of buildings using a regression ML model. Lastly, Paper III utilizes the building inventory dataset generated in Paper II to develop a material stock and flow model that introduces a new layered based approach to model renovation of buildings.

The findings presented in the appended work show that ML models can be used to predict physical attributes of roads and buildings to a high level of accuracy. The contribution of this work is showing that urban form features that can be generated using solely the geometry of roads and buildings can reliably achieve high level of prediction accuracy. Thereby increasing the applicability of the approach. The results also indicate that road maintenance and building renovation account for the largest share of embodied emissions. As a consequence, additional policy measures are needed to limit the emissions from maintenance and renovation activities.