Abstract:

This study aimed to establish a universal quantitative prediction model for soluble solids content (SSC) in
different varieties of grapes during storage based on visible and near-infrared diffuse reflection spectra. The mixed spectra
of Muscat, Manai and RedGloble grapes stored at 10 ℃ were taken as calibration materials, the influences of different
stoichiometrical calibration methods, spectral pretreatment methods, gaps, smooth points, different effective wavelength
intervals on the applicability of the established model for different grape varieties were examined. The results showed that
the modified partial least squares combined with 16 smoothing points, second derivative within 16 gaps and the scattering
method could produce the optimal model within the wavelength range of 408–1 092.8 nm with standard error of crossvalidation
(SECV) and coefficient of determination of cross-validation (R² _CV) of 0.308 7 and 0.980 2, respectively. The
model was evaluated via the prediction set of the above three varieties of grapes. The standard error of prediction (SEP) was
0.354, the correlation coefficient R_p² was 0.980 8, the relative prediction deviation (RPD) was 6.22, and the predicted residual
sum of squares (PRESS) was 7.993. When being applied for predicting the single varieties, the R_p² reached more than 0.94. Therefore, the near infrared detection model is useful to predict soluble solids content in grape and is suitable for different
grape varieties at the same time.

Key words: visible-near infrared diffuse reflectance spectroscopy, grape, storage, soluble solids content, prediction model