The Repeatability File—A Tool for Reducing the Sensitivity of near Infrared Spectroscopy Calibrations to Moisture Variation

Abstract

The study presented here demonstrates the benefits of using a repeatability file for near infrared (NIR) spectroscopy calibration development. Calibrations were developed from rapeseed samples of a narrow moisture content (2.5–6.0%) and used on a validation set with an extended moisture content (3.6–13.0%). The elevated moisture content of the validation samples led to increased standard errors of prediction ( SEP) for the calibrations predicting oil, protein, GSL and C18:1 content. Two methods were investigated to reduce the moisture sensitivity of the NIR spectroscopy calibrations: (a) the exclusion of water bands in the NIR spectral region during calibration development and (b) the use of a repeatability file as proposed by Westerhaus (1990). The increased SEPs could be decreased by up to 66% by either the exclusion of the water bands or the use of a repeatability file containing spectra of moist and dry samples. The best results were obtained with a repeatability file using spectra from a single sample. A spectral interpretation is given.