Near-infrared spectroscopy modeling for cellulose quantification in multiple species of plant biomass feedstocks

• A total of ten species of plant biomass feedstocks were used for model development. • SNV and CARS were the best spectral pretreatment and wavelength selection methods. • The optimum model’s R v 2 , RMSE P , and RPD V were 0.77, 3.53 %, and 2.09, respectively. • One prediction model of cellulose content is feasible for multiple species of biomass. This study aimed to assess if the cellulose contents of multiple species of biomass feedstocks can be predicted with a single model based on near-infrared spectroscopy (NIRS). We built a cellulose content prediction model for multiple species of biomass feedstocks, including cassava stem, maize leaf, maize stem, maize straw, peanut shell, Pueraria lobata stem, reed, rice straw, sorghum straw, soybean straw, switchgrass, and wheat straw samples. First, the best pretreatment method of NIRS was determined from nine pretreatment methods (including the original spectrum). Then, combined with partial least squares (PLS), five characteristic wavelength selection algorithms (including full wavelength) were used to establish models for the prediction of cellulose content. The performance of SNV-CARS-PLS was superior for quantifying the cellulose content. The root mean square error of prediction (RMSE P , %), the ratio of the standard deviation of the validation set to the standard error of prediction (RPD V ), and the coefficient of determination ( R v 2 ) were 3.53, 2.09, and 0.77, respectively. This study demonstrated that it was feasible to establish one model to predict cellulose contents of multiple species of biomass feedstocks using NIRS and an appropriate chemometrics method.