Abstract
Despite its potential, infrared spectroscopy combined with multivariate statistics has been seldom used to model peat properties with environmental value, such as the concentration of potentially toxic metals. In this research, we applied attenuated total reflectance (ATR) Fourier-Transform Infrared (FTIR) spectroscopy to evaluate the ability of the technique to predict mercury concentrations in late-Pleistocene/Holocene peat from a minerogenic peatland in Minas Gerais (Brazil). Mercury concentrations were analysed using a Milestone DMA-80 analyser and attenuated total reflectance FTIR-ATR was performed using a Gladi-ATR (Pike Technologies) in the mid-IR spectrum (4000–400 cm−1). Concentrations were modelled using principal components regression (PCR) and partial least squares regression (PLS). The performance of the models varied between moderate and very good (R2 0.67–0.90), with low RMSD values (0.35–1.06). A PLS model based on three latent vectors (LV1 to LV3) provided the best (R2 0.90, RMSD 0.35) results. LV1 reflected total organic matter content versus mineral matter (mainly quartz from local fluxes), LV2 was related to dust deposition from regional sources, and LV3 reflected peat organic matter decomposition. Compared to a previous investigation based on geochemical data, the spectroscopy-based PLS model performed better, but it has to be complemented with additional data (such as δ13C ratios) to reliably reproduce the changes of the factors controlling mercury accumulation over time. This time- and cost-effective methodology may help to develop multi-core approaches to study the within- and between-mire (of a similar type and area) variability in mercury accumulation, and probably also other peat properties.
