Improving the transfer of near infrared prediction models by orthogonal methods

Eight calibration transfer methods based on the removal of orthogonal signal were compared for the standardization of whole soybean protein and oil models. Dynamic orthogonal projection (DOP), transfer by orthogonal projection (TOP), error removal by orthogonal subtraction (EROS), orthogonal signal correction (OSC), and orthogonal projections to latent structures (O-PLS) as well as the modification and extension of some of these methods were compared in the transfer of models in intra and inter-brand situations using two Foss Infratecs and two Bruins OmegAnalyzerGs. For each brand, a master was designated and its models transferred onto the second unit of its network and the two units of the second brand. Calibration models were transferable from brand to brand with similar or better precision than when each instrument was calibrated on its own calibration set (for Infratec 1229, the relative predictive determinant (RPD) increased in intra and inter-brand calibration transfer situations from 10.42 to 11.45 and 10.57, with DOP and EROS respectively). Performance of each method was different across parameters, instruments, and validation sets. DOP modifications on the determination of the difference matrix showed promising results while TOP and EROS extensions to include variability specifically present in certain crop years did not bring any beneficial effects.