Abstract: An enhanced UV spectrometric method is applied to trace measurements of aromatic hydrocarbons dissolved in water. This approach gains selectivity and sensitivity by the use of optically generated first and second derivatives of transmission spectra. The augmented spectroscopic technique is combined with chemometric algorithms like principal component regression or partial least squares which are used for calibration of the spectrometer and quantitative evaluation of spectra. Laboratory measurements were performed on mixtures containing up to five substances, i.e. benzene, toluene, ethylbenzene, the three xylene isomers, chlorobenzene, and gasoline. Due to the difficulty of preparing precisely defined calibration and test samples of these very volatile compounds, a novel mixing device was developed and is presented. From these first investigations it can be estimated that the detection limits are down to ~10 µg analyte per liter of water by using a 10 cm absorption pathlength and a few minutes measurement time.

Abstract: A flow injection spectrophotometric method with multivariate calibration for quantification of BTEX compounds [benzene, toluene (I), ethylbenzene (II) and o-xylene (III) in effluents is investigated. Sample (200 µL), spiked with appropriate amounts of I, II and III was injected into a carrier stream of water (2.3 ml/min) of a flow injection system (schematic shown) and mixed with a stream of hexane (0.36 ml/min). The merged streams were carried to a PTFE microporous membrane solvent extraction cell and the extracted hexane layer was pumped to a silica flow cell (18 µL; 1 cm path length). The absorbance of the solution was measured at 2 nm intervals over the range 200-300 nm vs. a water reference. First- (FD) and second-derivative (SD) spectra were calculated with three-point wavelength smoothing, using the Savitsky-Golay algorithm. The absorbance of FD and SD spectra at 206 mm was linear up to 20 mg/l of pure I, II and III. The FD and SD spectral data were modeled using stepwise multiple linear regression, principal components regression and partial least-squares regression. Predictions of individual analyte concentrations in test samples produced prediction errors ranging from 8.5% for III to 45.8% for I. The lowest prediction errors (7%) were produced when calibrating the three compounds as a single component.