Abstract

Pretreated lignocellulosic fibers were used as a case study to compare two chemometric methods for the quantification of chemical components in Fourier transformed infrared (FT-IR) images. Partial least squares (PLS) and multivariate curve resolution with alternating least squares (MCR-ALS) methods were applied to the images to quantify glucans, lignin and hemicellulose content. The main problem for calibration in samples from natural origin is to obtain proper reference material for pixel to pixel quantification. Furthermore, chemical components in wood experience changes after different pretreatment conditions; therefore commercially available reference material may not have the same identity of the components present in the sample. Concentration information of bulk samples obtained by wet chemistry methods, along with the median spectrum of whole images, was used as an alternative for PLS calibration in this scenario. Results show that both methods provided similar spatial distribution for lignin and hemicellulose in the concentration maps, but image reconstruction of glucans shows differences in distribution between the two methods. PLS models used to quantify pixels in an image were previously validated through the prediction of global concentration of samples, using the median spectrum of different images (RMSEP = 1.3% for glucans, 1.0% for lignin and 0.9% for hemicelluloses); The range of pixel concentration predicted in a single image was too narrow possibly due to the lack of a calibration set with a wider dynamic range. Concentration maps obtained with MCR-ALS were satisfactory and the range of concentration for pixels was more consistent with what would be expected. A quantification approach that does not need a calibration set was used to transform concentration profiles into real concentration units for pixels. Therefore MCRALS was a more suitable method for quantification in this specific case study.