More accurate prediction results were found when using
calibration models built from combined data constituting two
years. Model performance was improved due to the wider range of
data and greater deviation of spectral features, which were
represented in the overall model. Lowest bias prediction results
were obtained from the overall three-year model. Acceptable
prediction capabilities were found (R2 > 0.70) with the best
performance being achieved for prediction of data for all
parameters from 2012 (Fig. 5). This can be explained by the fact
that fruits harvested in 2012 had the widest range of compositional
values, and thus, a more variable data set. The results corroborate
that model robustness can be improved by including a larger range
of data (Bobelyn et al., 2010; Golic and Walsh, 2006; Guthrie et al.,
2005; Louw and Theron, 2010; McGlone and Kawano, 1998; Peirs