By contrast, the models obtained for predicting juice weight and
juice content, both of which are of great importance to the industry,
must be deemed unacceptable in view of the results obtained
both at calibration (Table 3) and at external validation (Fig. 1).
Results for chemical parameters in intact mandarins are shown
in Table 3. The predictive capacity of models for pH and maturity
index may be considered acceptable (r2 = 0.75 and 0.74;
SECV = 0.09 and 1.16), whilst in terms of the recommendations
made by Williams (2001), the precision of models for predicting
soluble solids content and titratable acidity discriminated between
high, medium and low values, with r2 values of 0.55 and 0.64 for
SSC and TA, respectively.
The predictive capacity of the SSC model was poorer than that
reported for intact mandarins by Hernández-Gómez et al. (2006)
(RPD = 4.00; CV = 1.67%) and by Liu et al. (2010) (RPD = 2.90;
CV = 4.17%), although these authors used a monochromator in
the 350–2500 nm range and a diode-array instrument in the
600–980 nm range, respectively. Values of the statistics for the
models constructed in this study were better than those reported
by Hernández-Gómez et al., 2006 for pH (RPD = 1.73; CV = 3.25%)
and by Liu et al. (2010) for TA (RPD = 0.8; CV = 22.72%).
No published studies have addressed the calculation of the
SSC:TA ratio using NIRS technology in intact mandarins.
Validation statistics for the prediction of internal physical–
chemical quality parameters in intact mandarins both on-tree
and at harvest are shown in Fig. 1.
The model constructed for predicting pH in intact mandarins
using a portable MEMS-NIR instrument met the validation requirements
in terms of r2 and slope (r2 > 0.6; slope > 0.90) and both the
SEP(c) and the bias were within confidence limits: the equation
thus ensures accurate prediction, and can be applied routinely.
For pericarp thickness, maximum penetration force and the other
internal chemical quality parameters studied, it should be stressed
that SEP(c) and bias lay within confidence limits, although r2 and
slope results did not always attain recommended minimum values,
indicating that the NIRS equations constructed here can be considered
as a first step in the fine-tuning of NIRS technology for the ontree
monitoring of internal quality parameters in mandarins.