Validation of short wave near infrared calibration models for the quality and ripening of 'Newhall' orange on tree across years and orchards

The aim of this research was to test the viability of short wave near infrared spectroscopy (SW-NIRS) for the monitoring of fruit quality and ripening evolution in Algarve Citrus orchards (Citrus sinensis L. Osbeck 'Newhall'). Specifically, we have investigated the robustness of SW-NIRS calibration models in real conditions, that is: i) measurements were performed on tree, at a single location in the fruit equator, in the sunlight and with no temperature equilibration; and ii) with validation through independent data obtained in different years and/or orchards. Calibration models for soluble solids content (SSC), juice pH, titratable acidity (TA), firmness and maturation index (MI = SSC/TA) were built from the spectral data obtained in two orchards with different edaphoclimatic conditions, and in two consecutive years, corresponding to four independent datasets. We propose a method to assess model robustness through the comparison of internal validation (IV: calibration and validation data sets homogeneously sampled from the whole data set) and external validation (EV: calibration and validation data sets corresponding to different orchards and/or years). The method is based on the statistics of the results obtained by either IV and EV when applied to all the possible combinations of the four datasets. The results show that IV overestimates the models' performance relatively to the realistic exercise of EV. Globally, SSC and juice pH were the best performing models, with fair performances in IV and poor performances in EV (example for SSC: (IV/EV): rmsep = 1.00/1.15%, SDR = 1.40/1.13, R2 = 0.49/0.34). Firmness yielded the worse models, while TA and MI yielded intermediate performances. However, the plots derived from the comparison method suggest a convergence of IV and EV performances for larger numbers of samples, and thus the potential for future continuous model improvement.