Optimisation of the spectral pre-treatments used for Iberian pig fat NIR calibrations

The development of rapid analysis systems for the determination of the fatty acid profile of Iberian pig fat is decisive for the industries involved. Near infrared spectroscopy (NIRS) can fulfil the requirements of this sector, however NIR spectra of fat and oils can be subjected to uncontrolled error sources (changes in the instrument, environment, sample preparation procedure, etc.). Moreover, these errors are particularly important for the calibrations developed for these products, as they present sharp and narrow spectral peaks. There is evidence for correction of the effect of some spectral variations by the use of certain mathematical pre-treatments during calibration. Therefore, this work developed an automatic routine to obtain fatty acid calibrations (palmitic, stearic, oleic and linoleic acid) for Iberian pig fat, using 49 combinations of pre-treatments (first and second derivatives, Auto-Scaling, Detrending and two versions of Multiplicative Scatter Correction), in order to evaluate statistically the effects of different spectral corrections on the prediction ability of NIR equations affected by unexpected variations. Despite that the performance of prediction models showed variations with the mathematical pre-treatment used, generalised significant differences were not detected with the tests applied, except for linoleic acid. However, the optimisation of the derivatives used (order, gap and smoothing segment) proved to be essential for the implementation of robust calibrations.