Spatiotemporal estimation of construction and demolition waste generation using novel integrated machine learning and remote sensing approaches: A study of 83 Chinese cities

A widely recognized challenge hindering effective waste management planning worldwide is a lack of systematic and continuous data on construction and demolition waste (CDW) generation. To address these critical data limitations, this study presents its first systematic effort to integrate remote sensing (RS) data into a machine learning (ML)-based framework for estimating CDW generation. Utilizing all currently available data collected from selected Chinese cities, four widely used ML algorithms were trained and evaluated: Random Forest (RF), Gradient Boosting Decision Trees (GBDT), Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (XGBoost) using three feature groups: statistical variables, RS variables, and a combination of both. The best-performing models were applied to an interpolation dataset covering 83 Chinese cities over a 10-year period to evaluate their spatiotemporal extrapolation capabilities and generalizability. The results consistently indicate that models based solely on RS features outperform those using only statistical data or combined feature sets, with R² values ranging from 0.77 to 0.81 across all four algorithms. Among these, the RF model exhibited the highest overall performance, while LightGBM and XGBoost also delivered competitive results. Analysis of CDW generation across 83 Chinese cities revealed distinct spatial hotspots in regions such as the Yangtze River Delta and the Pearl River Delta, with a noticeable inland expansion trend over time, reflecting China's national strategies for promoting balanced regional development. This study offers a novel, scalable, and transferable approach that expands the methodological boundary of CDW estimation beyond conventional statistical data, offering new insights and practical implications for global CDW management, particularly in rapidly urbanizing or data-scarce regions.