The calibration statistics for the prediction of physical properties
(firmness, maximum penetration force, pericarp thickness,
juice weight and juice content) and chemical parameters (soluble
solids content, pH, titratable acidity and maturity index) in intact
mandarins are shown in Table 3.
The best equations for predicting physical–chemical parameters
were obtained using SNV + DT as scatter-correction treatments,
though not in the case of SSC; this underlines the importance of
selecting spectral-signal pretreatments as a function of the parameter
to be analyzed (Delwiche and Reeves, 2004).
The models displaying the greatest predictive capacity were
constructed using the second derivative of the spectrum, except
for the parameters firmness, juice weight and SSC, where the first
derivative provided better results.
With regard to texture-related parameters, models obtained for
maximum penetration force displayed greater predictive capacity
than those constructed for firmness (Table 3). The poor predictive
capacity of firmness models highlights the difficulty in correlating
destructive measurements made to a puncturing depth of 10 mm
with non-destructive NIR measurements. As Peirs et al. (2002)
have noted, NIR light will only penetrate usefully down to a depth
of between 1 and 5 mm, depending on the wavelength, the instrument
and the fruit ripeness stage. For maximum penetration force,
the value of the coefficient of determination (r2) (0.47) for the best
model obtained suggests that this model would only enable discrimination
between high and low maximum penetration force
(Williams, 2001).
These findings agree with those reported by Sánchez et al.
(2011), who stress the difficulty in predicting texture-related
parameters (firmness and maximum penetration force) in certain
fruits using NIRS technology with MPLS regression.
No references have been found in the literature to the measurement
of these parameters in intact mandarins using NIRS technology.
Hernández-Gómez et al. (2006) measured compression force
(i.e. the force required to compress a fruit by 3% of its diameter)
in intact Satsuma mandarins, using a monochromator instrument
with a spectral range of 350–2500 nm, obtaining a reasonable-togood
prediction performance (r2 = 0.75; RPD = 1.77). However, this
texture parameter was not measured in the present study, so findings
cannot be compared.
The predictive capacity of the best model for predicting pericarp
thickness may be considered acceptable for screening purposes,
since the value recorded for the coefficient of determination for
cross-validation (r2 = 0.52) would enable samples to be classified
as high, medium or low; given the link between pericarp thickness
and fruit yield, this classification would be of considerable value to
the citrus-fruit sector.