Correlation of limestone beds is commonly based on a variety of features, including the age of the bed, the fossil assemblage, internal sedimentary structures, and the relationship to other units in the stratigraphy. This study uses laser-induced breakdown spectroscopy (LIBS) to correlate 16 limestone beds from Kansas, USA, using three multivariate techniques: (1) soft independent modeling of class analogy (SIMCA) classification, (2) a partial least squares regression, 1 variable (PLS-1) model in which the spectra are regressed against a matrix of the indicator variables 1 through 16, and (3) a matching algorithm that consists of a sequence of binary PLS-1 models. Each gravel-sized limestone particle was analyzed by one LIBS shot; ten spectra were averaged into a single spectrum for chemometric analysis. The entire spectrum (198–969 nm wavelength) is used for multivariate analysis; the only preprocessing is averaging. The SIMCA and PLS-1 models fail to discriminate among the beds, which are chemically similar. In contrast, the matching algorithm has a success rate of 95% to 96%, using half of the spectra to train the model and the other half of the spectra to validate it. However, 100% success can be accomplished by accepting the classification of the majority of spectra for a given bed as the correct classification. This study indicates that LIBS can be applied to complex geologic correlation problems and provide rapid, accurate results.
© 2012 Optical Society of AmericaFull Article | PDF Article