Laser induced break down spectro-analysis an d characterization of environmental matrices utilizing multivariate chemo metrics

Abstract

We exploited multivariate chemometric methods to reduce the spectral complexity and to retrieve heavy trace metal analyte signatures directly from the LIBS spectra as well as to extract latent trace meta l profile characteristics in two important envi ronmental samples (soils, rocks) sampled from a geothermal field associated with a high background radiation area (HBRA). As, Cr, Cu, P band Ti were modeled for direct trace (quantitative) analysis using partial least squares (PLS) and artificial neural networks (ANN) with regression coefficients ( R2 ) ~ 0. 99. PLS performed better in soils than in rocks; the use of ANN improved the accuracies in rocks because ANNs are more robust than PLS at correcting matrix effects and modeling spectral non - linearities. The predicted trace metal profiles together with LI BS atomic and molecular signatures acquired using single ablation in the 200 – 5 45 nm spectral range were utilized to successfully classify and identify the soils and rocks with regard to whether they were derived from (i ) a high background radiation area (HB RA) - geothermal, (ii) HBRA - non - geothermal or (ii) non - HBRA - geothermal field using principal components analysis (PCA) and soft in dependent modeling of class analogy (SIMCA)