the model was very low (5%) and the RMSECV (0.20 mg g−1 fat) was
higher than the SD (0.18 mg g−1 fat), generating a RPD lower than 1.0.
Again, a narrower range of variability for this FA together with a low
concentration could have negatively influenced the NIRS prediction.
When looking at the equation predictions performed with the NIR
spectra collected on cold samples, the accuracy of prediction was lower
for n−3, C18:3n−3 and C20:3n−3 (R2=0.77–0.80; RMSECV=0.12–
1.75 mg g−1 fat; RPD=1.76–1.83). During the trial itwas observed that
when the samples were warmed to 31 °C, the fat which occasionally
showed small and unremovable air bubbles became free of these
bubbles and also became slightly translucent. A less homogeneous
distribution of fat throughout the cells and more air bubbles or reduced
molecular vibration due to the cooler temperature could have been the
reasons for the poorer predictions when using cold samples. Thus, NIR
spectroscopy showed a higher predictability of estimation for n−3 FA
content on intact warm than on cold samples. This could be useful for
early in-plant identification of beef fat that is enriched with these FA.
Regarding the n−6/n−3 ratio, the NIRS predictability was low when
both warm and cold samples were scanned (R2=0.71 and 0.74;
RMSECV=0.98 and 1.07 mg g−1 fat; RPD=1.51 and 1.44; respectively).