Jp. Conzen et al., DETERMINATION OF CHLORINATED HYDROCARBONS IN WATER BY FIBEROPTIC EVANESCENT-WAVE SPECTROSCOPY AND PARTIAL LEAST-SQUARES REGRESSION, Fresenius' journal of analytical chemistry, 348(8-9), 1994, pp. 501-505
An application of the multivariate calibration technique of partial le
ast-squares (PLS) regression to near-infrared spectra of a fiber-optic
sensor based on the evanescent wave principle is presented. The sensi
ng element consists of a quartz glass fiber with a silicone cladding w
hich enriches nonpolar water contaminants. Due to the interaction of t
he extracted molecules with the part of the light which is transmitted
in the evanescent wave zone of the cladding, absorbance spectra of th
e contaminants can be collected. In view of a sensor application for i
n-situ environmental analysis, aqueous solutions of chlorinated hydroc
arbon solvents (CHS), which often can be found as major water contamin
ants, have been measured. PLS regression was applied to three sets of
CHS samples, representing typical features of NIR evanescent wave spec
tral data. These are, e.g., strong overlapping of the absorption bands
of different CHS components, peak distortions due to temperature vari
ations between reference and sample measurement and noisy data at anal
yte concentrations near to the limit of detection, respectively. For t
richloroethene and 1,1-dichloroethene, where the calibration model was
built for samples within a small concentration range of 1-9 mg l-1, s
atisfactory prediction results could be obtained with a relatively sma
ll root-mean-square error of 0.3 mg l-1 compared to analytical referen
ce measurements. In contrast to this, for a three component system of
dichloromethane, trichloromethane and trichloroethene with strongly ov
erlapping absorption bands, where samples over a very broad concentrat
ion range from 3-4940 mg l-1 were included in the PLS model, the predi
ction accuracy decreased enormously and for some samples strong deviat
ions between real and predicted data occurred. Nevertheless, applying
multivariate calibration to this difficult system with similar spectra
l features and huge differences in the concentration of the species al
lowed an acceptable spectral distinction and at least a semi-quantitat
ive determination of the CHS species.