In the context of a circular economy, the built environment represents a vast future stock of recoverable resources. Anticipating when and where buildings will be demolished, combined with knowledge of their material composition, enables targeted reuse and recycling strategies and contributes to lifecycle carbon mitigation.
This study presents an integrated framework that combines building-level demolition forecasting with material composition prediction to map reclaimable materials across space and time. A consolidated dataset was created by integrating Geographic Information System (GIS) data with building attributes (e.g., age, height, type) and socio-economic indicators. Demolition likelihood was predicted using spatio-temporal deep learning models capable of capturing both the spatial context and temporal evolution of urban areas.
To estimate material stocks, building typologies were analysed and their component compositions extrapolated via machine learning models trained on building features from BIM models of existing buildings. By linking these two predictive layers, the framework generates spatio-temporal maps of reclaimable materials, identifying demolition hotspots and material quantities (e.g., concrete, steel, timber, brick) within the building stock.
This framework provides a scalable, data-driven approach for quantifying future secondary material flows and their geographic distribution. By integrating demolition forecasting with material composition prediction, it enables stakeholders to anticipate material availability, optimise recovery strategies, and plan for circular resource use in urban development. The approach will be demonstrated through a case study in Switzerland, with results highlighting potential material yields, spatial patterns of resource availability, and opportunities for targeted urban mining. Ultimately, this work supports the transition toward a low-carbon, resource-efficient built environment.