Satellite-based detection of asbestos-cement roofs using WorldView-3 VNIR data: an affordable evaluation tool for sustainable management of hazardous building infrastructure

Asbestos-cement (AC) roofing remains a critical concern for urban environmental health, particularly in cities where asbestos bans were implemented late and legacy materials persist. Reliable identification of AC roofs is essential for risk assessment, remediation planning, and policy enforcement. This study introduces a computationally efficient framework for AC roof detection using eight visible- and near-infrared (VNIR) bands from WorldView-3 imagery, reducing reliance on hyperspectral or SWIR data in existing approaches. Thirty-two supervised classifiers spanning eight methodological families were evaluated under both a five-class urban material scheme and a binary AC versus non-AC scenario. Nearest-neighbour and kernel-based approaches consistently outperformed alternative strategies. Fine K-Nearest Neighbours achieved the highest performance in the multiclass setting, reaching a Macro F1-Score of 97.6% on independent test data, while Subspace KNN proved robust for binary AC detection, attaining near-perfect discrimination with test accuracies approaching 99–100%. Feature attribution analysis confirmed that classification decisions are driven by reflectance contrasts in the red-edge and near-infrared regions, reinforcing the reliability of the results. The study provides one of the first comparative assessments of computational efficiency across classifiers for urban material mapping, showing that reducing class complexity improves processing speed without compromising detection reliability. The findings establish that VNIR-only satellite imagery with lightweight machine-learning models constitutes a viable solution for mapping asbestos-cement roofs, with direct relevance for urban planners, environmental agencies, and public health authorities seeking cost-effective tools for large-area asbestos monitoring. • VNIR WorldView-3 data offers a low-cost solution for asbestos roof detection. • Benchmarking of 32 algorithms identifies Fine and Subspace KNN as optimal models. • Binary classification schemes achieved >98% accuracy with high stability. • Red-Edge and NIR bands are sufficient for precise hazardous material mapping. • Results validate a cost-effective method for mapping asbestos in developing countries.