Quantitative characterization of lignocellulosic biomass using surrogate mixtures and multivariate techniques

PLS regression models were developed using mixtures of cellulose, xylan, and lignin in a ternary mixture experimental design for multivariate model calibration. Mid-infrared spectra of these representative samples were recorded using Attenuated Total Reflectance (ATR) Fourier Transform Infrared (FT-IR) spectroscopy and regressed against their known composition using Partial Least Squares (PLSs) multivariate techniques. The regression models were cross-validated and then used to predict the unknown compositions of two Arabidopsis cultivars, B10 and C10. The effect of various data preprocessing techniques on the final predictive ability of the PLS regression models was also evaluated. The predicted compositions of B10 and C10 by the PLS regression model after second derivative data preprocessing were similar to the results provided by a third-party analysis. This study suggests that mixture designs could be used as calibration standards in PLS regression for the compositional analysis of lignocellulosic materials if the infrared data is appropriately preprocessed.