Influence of Yearly Variability of Agricultural Products on Calibration Process: A Triticale Example

Many elements can influence the calibration process for near-infrared spectroscopy: variability within the population of interest, environmental effects, hardware instability, and others. This study evaluated two techniques to develop prediction models for triticale grain moisture and protein. The spectral addition strategy added new samples to the calibration set year after year. The spectral adaptation strategy selected only the spectral variability needed to successfully apply the model to new material. Triticale was a good study material because it was undergoing genetic change through the study period and is very responsive to climate variability. The two calibration techniques were significantly different from each other in terms of precision and accuracy. Spectral adaptation was the best technique with a relative predictive determinant (RPD) of 4.33 and a bias of 0.17% versus RPD of 3.50 and a bias of –0.52% for spectral addition. These results are contradictory to common practices that tend to add to the calibration set a maximum variability over as much time as possible. For highly variable matrixes, a constant adaptation rather than expansion of the calibration pool may be more appropriate.