What matters most to the material intensity coefficient of buildings? Random forest‐based evidence from China

Abstract

Material intensity coefficient (MIC) is vital for material stock accounting in the field of industrial ecology. However, the categorization of MIC varies across regions especially for buildings that diverge greatly along the history and space aspect, and acquisition of MIC data and building information have always been a challenge in related studies. In this study, the state-of-art ensemble model "Random Forest" was developed on Chinese buildings to identify the impact of four building attributes (building structure, construction year, use type, and region) on MIC, and these features' importance was further assessed by considering variable correlations. The features' importance and their individual effects on MIC were intuitively revealed by depicting the partial dependence plots. Finally, a set of hierarchical MIC values was estimated by integrating the order of four variables' importance and a quick MIC calculator was provided. Resultsshowed that building structure is the most influential attribute for MIC, followed by the construction year, use type, and region, successively. The RF-based MIC values allow researchers to apply it to material stock and flow analysis by choosing a specific building feature(s) in the MIC calculator, which is (are) available in building physical inventory data. This study provides a method that could help researchers locate key influencing variables and give insights into the comparability of MIC research across regions and play an important role in developing urban mining and circular economy strategies.

Gridded material stocks in China based on geographical and geometric configurations of the built-environment

“…Buildings in urban and rural areas were categorized into urban residential buildings (UR), urban non-residential buildings (UnR), rural residential buildings (RR), and rural non-residential buildings (RnR) 20 – 24 . Firstly, to demarcate material stocks between rural and urban material stocks, the GHSL settlement layers (version of GHS-SMOD R2022A) 25 , namely an open-data project providing global spatial information about the settlement classification over time, were employed to delineate urban and rural buildings geographically.…”

RASMI: Global ranges of building material intensities differentiated by region, structure, and function

“…Many of these challenges are further described in the comparative MI studies cited above 48 – 51 and in recent reviews of material stocks research 7 , 8 .…”

“…Moving beyond descriptive statistics, Zhang et al . 51 assessed the variability of MIs in the Chinese database, using a machine learning approach. Nasiri et al .…”