Assessment and reduction of embodied carbon emissions in buildings: a systematic literature review of recent advances

Decarbonization of the building sector is crucial for achieving global carbon neutrality, as the industry constitutes 37 % of total greenhouse gas (GHG) emissions worldwide. While operational carbon emissions (OCE) have declined through energy efficiency advancements, embodied carbon emissions (ECE) remain challenging to address due to their complexity across multiple building lifecycle stages. Conventional Life Cycle Assessment (LCA) faces limitations in dynamic scenarios, regional variability, and system boundary inconsistencies. Following the PRISMA 2020 guidelines, we conducted a systematic review of 7,731 publications (2020–2024), ultimately analyzing 159 high-impact studies. This review employs bibliometric analysis and qualitative content analysis to identify methodological innovations and effective reduction strategies. Bibliometric analysis revealed a notable shift from traditional LCA approaches toward two pivotal paradigm shifts: (1) machine learning (ML)-driven methodologies for dynamic assessment and reduction, which leverage three fundamental learning paradigms (supervised, unsupervised, and reinforcement learning) to automate predictive modeling, uncertainty quantification, and design optimization across the building lifecycle; and (2) multi-criteria decision-making (MCDM)-based frameworks that integrate regional-specific variables to resolve the carbon-performance-cost trilemma. Case studies from diverse geographical contexts highlight the necessity of context-sensitive strategies, demonstrating how localized decarbonization pathways can reduce ECE while maintaining building performance. Key challenges include data heterogeneity and standardization gaps, interpretability limitations of advanced ML algorithms, and geospatial adaptation of MCDM frameworks. Key findings emphasize the need for hybrid ML-MCDM frameworks with enhanced explainability that deliver actionable insights for ECE reduction in diverse contexts. Future research should prioritize the development of interpretable ML models that balance computational efficiency with transparent design parameter justification, establishing unified protocols for ECE accounting across diverse contexts, and fostering cross-disciplinary partnerships.