Spatial determinations and application of land-use regression models to evaluate outdoor volatile organic compounds in Taiwan

Volatile organic compounds (VOCs) emitted from stationary and mobile sources cause public health concerns; however, a nation-scale model for detailed exposure assessment of the entire population in Asian is lack. This study conducted an eight-month sampling campaign to evaluate outdoor VOC concentrations and establish land-use regression (LUR) models for specific VOC components in Taiwan. A total of 31 district offices were randomly selected from 349 administrative centers to perform a 24-h sampling campaign for seven VOC components: benzene (B), toluene (T), ethylbenzene (E), xylene (X), styrene (S), dichloromethane (DCM), and n-hexane (n–H). Road traffic information, meteorological data, and points of interest were collected and combined with a geographic information system to establish LUR models for specific VOC components. The median concentrations of B, T, E, X, S, DCM, and n–H were 0.95 μg/m³ (range: 0.85–7.96 μg/m³), 1.21 (1.13–7.00) μg/m³, 0.59 (0.53–5.21) μg/m³, 2.01 (0.88–6.86) μg/m³, 0.82 (0.76–5.52) μg/m³, 43.90 (2.44–488.22) μg/m³, and 0.60 (0.55–5.11) μg/m³, respectively. DCM was identified as the highest component compared with other VOC species. The predictive capacities (R²) for specific LUR models were 0.71, 0.36, 0.60, 0.63, 0.41, 0.68, and 0.82. The differences between the model R² value and leave-one-out cross-validation R² value ranged from 1 to 7% for seven VOC components. This study established LUR models with good predictive capabilities to estimate B, E, X, DCM, and n–H in Taiwan. These built models can be applied for exposure assessment of specific VOC components in environmental epidemiological studies.