Embodied greenhouse gas emissions of buildings—Machine learning approach for early stage prediction

Abstract

Observations made in architecture and engineering practices have highlighted the need to access buildings embodied greenhouse gas (GHG) emissions at early stages as well as a general understanding of the impacts of design decisions. This research addresses this need, and lack of readily available methods by leveraging knowledge from existing building’s databases, and using contextual and descriptive data, such as structure typology and location, to quantify the total embodied GHG emissions. It uses machine learning to develop a methodology for early and hands-on approximation of embodied GHG emissions, allowing to compare buildings, explain feature impacts, and verify computed results. The methodology, tested on the Embodied Carbon of European Buildings database, is generic and can be trained on other building databases and provide predictions tailored to their content. Alongside direct applications for design and decision-making, it provides a systematic analysis of features and emission standards, which, applied to a national database, could help inform policy models and mitigation strategies and transition towards a low emission and sustainable built environment.