Determinação de hidrocarbonetos policíclicos aromáticos utilizando calibração multivariada

Abstract

A method was developed to determine low molecular weight polycyclic aromatic hydrocarbons (naphthalene, anthracene, fluorantene, phenanthrene and acenaphthene) in aqueous leachates from soils and sediments using a salting-out assisted liquid-liquid extraction, synchronous fluorescence spectrometry and a multivariate calibration technique. Several experimental parameters were controlled and the optimum conditions were: sodium carbonate as the salting-out agent at concentration of 2 mol L-1, 3 mL of acetonitrile as extraction solvent, 6 mL of aqueous leachate, vortexing for 5 min and centrifuging at 4000 rpm for 5 min. The partial least squares calibration was optimized to the lowest values of root mean squared error and five latent variables were chosen for each of the targeted compounds. The regression coefficients for the true versus predicted concentrations were higher than 0.99. Figures of merit for the multivariate method were calculated, namely sensitivity, multivariate detection limit and multivariate quantification limit. The selectivity was also evaluated and other polycyclic aromatic hydrocarbons did not interfere in the analysis. Likewise, high performance liquid chromatography was used as a comparative methodology, and the regression analysis between the methods showed no statistical difference (t-test). The proposed methodology was applied to soils and sediments of a Brazilian river and the recoveries ranged from 74.3% to 105.8%. Overall, the proposed methodology was suitable for the targeted compounds, showing that the extraction method can be applied to spectrofluorometric analysis and that the multivariate calibration is also suitable for these compounds in leachates from real samples. Moreover, a proof of concept multivariate method to determine seven other polycyclic aromatic hydrocarbons (2-naphthol, hydroxypyrene, hydroxyfluorene, crysene, benzo(a)anthracene, benzo(b)fluoranthene and benzo(b)pyrene) was developed and its figure of merits were calculated. The built method had its prediction ability evaluated and it showed to be promising for application in real samples.