Enhanced Databases on City’s Building Material Stock. An Urban Mining Method Based on Machine Learning for Enabling Building’s Materials Reuse Strategies

The climate crisis and the growing urbanization needs urge design and construction practices to shift their focus to the anthroposphere as a source of, rather than just a destination for, building materials. The concept of urban mining is revisited by many to manage the existing building material stock exploring the potential for reuse in new constructions. By combining image-based segmentation with cadastral data in a GIS database, this paper proposes an end to end process for an integrated web-based application enclosing and delivering cross-referenced data about a city’s material stock. The study uses unstructured data from Open Street View to identify relevant patterns for making estimations of the quantity, state, and projected availability cycle of concrete, brick, stone, metal, timber, and glass in building facades. By applying predictive modelling at the city scale, the algorithm can identify, geolocate and quantify façade materials with a present accuracy of 87%. The developed method proposes an alternative automated way to physical inspections which makes it applicable to different cities that lack registers of building data. A data repository map is developed in the format of an interface, so that it can be used by different stakeholders including decision-makers in the formulation and planning of urban material reuse strategies, as well as designers in the early stage of circular design processes. Tackling the accuracy limitations of machine learning, the paper concludes with studying the potential of combining the use of image unstructured data with statistical data of existing building registers including age of buildings and preservation state.