Estimation of low concentrations in the presence of high concentrations using Bayesian algorithms for interpretation of spectrophotometric data

• Published in: Journal of chemometrics 2004-05, Vol.18 (5), p.217-230
• Main Authors: Niedziński, Cezary, Morawski, Roman Z.
• Format: Article
• Language: English
• Subjects: Algorithms; Bayesian analysis; Bayesian procedures; Chemistry; data processing; Exact sciences and technology; General and physical chemistry; least-squares estimation; monitoring; Parameter estimation; spectrochemical analysis; Spectrum analysis; UV-Vis-NIR spectroscopy
• ISSN: 0886-9383; 1099-128X
• DOI: 10.1002/cem.851

A Bayesian method for interpretation of spectral data, dedicated to the monitoring‐type applications of mini‐ and micro‐spectrometers, is addressed. The method generates estimates of the concentrations of the components of an analyzed substance on the basis of the data representative of its absorption spectrum, provided that both the normalized spectra of the components and statistical information on historical measurements of the monitored concentrations are available. The Bayesian method is systematically compared with the constrained least‐squares method, under an assumption that the estimated concentrations differ considerably and the processed data are subject not only to random instrumental noise but also to some random disturbances introduced by the residual content of components of the analyzed substance being not identified. A study, performed using both synthetic and real‐world spectrophotometric data, is aimed at the assessment of the robustness of the Bayesian method to the maximum‐to‐minimum ratio of concentrations and imprecision of a priori information. Copyright © 2004 John Wiley & Sons, Ltd.