NIR spectral information used to predict water content of pine seeds from multivariate calibration

It is possible to predict the water content (6%-32%) of pine seeds - single seeds at 30 degree-days and bulk samples at 45 degree-days - using multivariate calibration models based on the near infrared (NIR) spectra (1100-2200 nm) of the seeds. One would expect the water peaks in the NIR spectra to contribute uniquely to the models, but this is not entirely the case. Different ways of studying models and spectra reveal that the contribution of the spectral information to the calibration models is complicated and nonlinear. Model parameters also show contributions from the biological processes in the seeds. Regression models between water content and NIR spectra using biorthogonal partial least squares (BPLS) showed that water content was associated with overtones of H-O-H, mainly around the peak of water at 1930-1940 nm, as expected. However, overtones of C-H, C=O, and N-H also influenced the BPLS models. This was caused by evolving biological phenomena, such as respiration and protein metabolism in imbibed seeds, and hence gave more complex regression models of seed-water interaction. There was also a difference in N-H absorption that indicated enhanced protein metabolism at prolonged degree-days.