Answers are grounded in this paper and cite their sources.
Illegal landfilling of construction and demolition waste (CDW) creates a critical loophole in circular economy and net-zero transitions. Although remote sensing and deep learning have improved CDW-site detection, previous studies have mainly focused on mapping suspected waste locations, with limited attention to linking detected candidate sites with spatial conditions associated with their occurrence. This study develops a framework integrating high-resolution satellite imagery, ensemble deep learning, and interpretable machine learning. Using Scotland as a case study, U-Net, U-Net++, and DeepLabV3+ were trained to detect landfill-like surface disturbances, with soft-voting ensembles combining their predictions. Retained candidate sites were combined with remote sensing-derived environmental and socioeconomic indicators to construct a spatial analysis dataset. Random Forest, XGBoost, and LightGBM were compared for occurrence modelling, with Shapley additive explanations used to interpret the selected model. The U-Net++ and DeepLabV3+ ensemble performed best (test intersection over union, IoU = 0.6746). After spatial filtering, 43 potential illegal CDW landfill sites were retained across Edinburgh, Glasgow, and Aberdeen. XGBoost achieved the highest cross-validated area under the precision–recall curve and was selected for interpretation. The model explanations show that building density made the largest contribution to model predictions, followed by vegetation cover, distance to roads, crime rate, and income deprivation, with candidate sites mainly associated with low-density areas. By integrating site detection with interpretable spatial explanation, this study shows how spatial evidence can support a shift in illegal CDW landfill governance from reactive clean-up towards proactive, risk-informed monitoring, while providing a potential basis for circular construction, enforcement, and waste governance that may be adapted to other data and regulatory contexts.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: