Prediction of Chemical Characteristics of Fibrous Plant Biomasses from Their near Infrared Spectrum: Comparing Local versus Partial Least Square Models and Cross-Validation versus Independent Validations

Abstract

The reliability of local and partial least square (PLS) near infrared (NIR) models to predict the chemical characteristics of fibrous plant biomasses was compared. Validations with different degrees of independence were used. The developed NIR models were reliable for the prediction of different main chemical characteristics of various fibrous plant species using multispecies datasets. The local models were more reliable in terms of prediction error compared with the PLS models because the local method appears to cope with the nonlinearity and non-homogeneity associated with a large multispecies dataset. The degree of independence of samples in the validation set relative to samples used in the calibration set had a major impact on the prediction performance, especially for the local method. It affected the local method more because of the lower number of samples used in its specific regressions. There was a decrease in the reliability of local and PLS models according to the increase in the degree of independence of the validation set (i.e. the similarity of the predicted samples in regard to the calibration samples). The additions of a few independent samples of the predicted plant-species group to their calibration set that did not contain samples of the predicted plant-species group improved the prediction performance of multispecies models, especially for the local method. The type of NIR models developed in the present study can be used for screening, ranking and quantitative analyses of the main chemical components contents in fibrous biomasses, and for the assessment of their suitability to be converted into biofuels.