Search

The aim of this study was to determine the impact of incorporating switchgrass samples that have been in long term storage on the development of near-infrared (NIR) multivariate calibration models and their predictive capabilities. Stored material contains more variation in their respective spectral signatures due to chemical changes in the bales with storage time. Partial least squares (PLS) regression models constructed using NIR spectra of stored switchgrass possessed an instability that interfered with the correlation between the spectral data and measured chemical composition. The models were improved using calibration sample sets of equal parts stored and fresh switchgrass to more accurately predict the chemical composition of stored switchgrass. Acceptable correlation values (rcalibration) were obtained using a calibration sample set composed of 25 stored samples and 25 samples of fresh switchgrass for cellulose (0.91), hemicellulose (0.74), total carbohydrates (0.76), lignin (0.98), extractives (0.92), and ash (0.87). Increasing the calibration sample set to 100 samples of equal parts stored to senesced material resulted in statistically increased (p = 0.05) correlations for total carbohydrates (0.89) and ash (0.96). When these models were applied to a separate validation set (equal to 10% of the calibration sample set), high correlation coefficients (r) for predicted versus measured constituent content were observed for cellulose (0.94), total carbohydrates (0.98), lignin (0.91), extractives (0.97), and ash (0.90). For optimization of processing economics, the impact of feedstock storage must be investigated for implementation in conversion processes. While NIR is a well-known high-throughput technique for characterization of senesced switchgrass, the selection of appropriate calibration samples and consequent multivariate models must be taken into careful consideration for NIR application in a biomass storage facility for rapid chemical compositional determination. [ABSTRACT FROM AUTHOR]