Detection limit estimator for multivariate calibration by an extension of the IUPAC recommendations for univariate methods

Abstract

A methodology is proposed to estimate the limit of detection (LOD) of analytical methods when multivariate calibration is applied. It tries to follow the same premises as the IUPAC methodology for univariate calibration. The mathematical support is given and algorithms such as partial least squares (PLS) regression, PLS2 and principal component regression (PCR) are used. Only multivariate raw data are used; that is, no surrogate univariate signal is deduced. Non-linearities are allowed. Near infrared (NIR) data of 5 component pseudo-gasoline samples together with simulated fluorescence synchronous spectra of binary mixtures (first order data) are used for evaluation. Experimental verification is performed using different kinds of data, namely: binary mixtures of bentazone and fenamiphos (very overlapped spectra, second order data) obtained by sequential injection (SI), and kinetic data of the reaction between the Fenton's reagent (FR) and pesticides such as atrazine, bentazone and alachlor (individual or binary mixtures, second order data). Results are always compared with independent methods previously proposed in the literature, based in the use of surrogate univariate signals. In general, similar results are found and no statistically significant differences seem to be present, except in a few cases when complex chemical systems are involved.