Chemometric Tools for Quantitative FTIR Spectroscopy of Soils
Department of Soil Science, Faculty of Natural Sciences, Comenius University, Bratislava
Keywords: FTIR spectroscopy, soil, quantitative analysis, multivariate statistical calibration, principal component regression, partial least squares regression, artificial neural networks
The potential of FTIR spectroscopy to provide reliable quantitative data on soils has been explored intensively within last decade and a number of works dealing with this subject have been published. In these studies (similarly as in other areas of chemical analysis), calibration of FTIR data was performed mainly by use of multivariate statistical techniques. In this context, principal component regression (PCR) or partial least squares regression (PLS) has been frequently employed. In some studies, however, more advanced calibration models, such as artificial neural networks (ANN) have been used. The aim of the present contribution was to summarize basic knowledge of multivariate calibration techniques used in quantitative FTIR spectroscopy of soils. It reviews fundamental aspects of quantitative measurement using FTIR, characterizes the principles of the preliminary data exploration, and discusses pre-processing possibilities of collected spectra. A substantial part of the text deals with the PCR and PLS techniques, the use of which is characteristic for the FTIR calibration. Besides PCR and PLS, the theory of ANN is briefly discussed and the example of its use is included. Particular methods which serve for accuracy assessment of calibration models are presented as well.
Full text (PDF)