Multi-product calibration models using NIR spectroscopy

Abstract

The physical–chemical composition of multiple biomasses can be predicted from one single calibration model instead of compositional prediction conducted by individual models. In this work, multi-product models, involving banana, coffee and coconut samples were built by partial least square regression (PLS) for ten different chemical constituents (total lignin, klason lignin, acid insoluble lignin, acid soluble lignin, extractives, moisture, ash, glucose, xylose and total sugars). The developed PLS models show satisfactory results, with relative error (RE%) less than 20.00, except for ash and xylose models; ratio performance deviation (RPD) values above 4.4 and range error ratio (RER) values above 4.00. This means that all models are qualified for screening calibration. Principal component analysis (PCA) was useful to demonstrate the possibility and the rationale for combining three biomass residues into one calibration model. The results have shown the potential of NIR in combination with chemometrics to quantify the chemical composition of feedstocks.