Affordable risk mapping and detection of asbestos-cement roofs via remote sensing: towards accessible urban digital twins in low- to middle-income countries

Deteriorating asbestos-cement (AC) roofs in urban areas pose environmental and public health risks from carcinogenic asbestos fibers. Accurate identification of these structures is crucial for hazard assessment and mitigation strategies. This study compared four supervised classification algorithms for AC roof detection using multispectral remote sensing data: Minimum Distance (MiD), Mahalanobis Distance (MhD), Spectral Angle Mapper (SAM), and Support Vector Machine (SVM). The dataset includes RGB, four- and eight-band multispectral, and hyperspectral images from an urban test site. Using ENVI® software, each classifier's performance was assessed across five land cover classes: AC, steel roofs, clay roofs, roads, and vegetation. A key novelty is the comparative analysis of these methods across a multi-scale dataset. High spatial resolution (0.18 m RGB and four-band imagery) achieved overall accuracies of 95% and 96%, respectively. Results indicate that readily available optical data with high spatial detail can outperform spectrally richer but coarser-resolution (1.2 m) satellite data in accuracy. This study emphasizes the discriminatory power of visible-near-infrared (VNIR) spectral information for AC identification across image types, contrasting with literature that relies on costly Shortwave Infrared (SWIR) data. This finding suggests a more accessible pathway for AC roof detection, beneficial for low- and middle-income countries transitioning from asbestos. A conceptual digital twin framework for urban roofing is proposed, offering authorities a tool to monitor AC roof remediation and curb illicit removal in low-to middle-income economies.