Development of predictive models for quality and maturation stage attributes of wine grapes using vis-nir reflectance spectroscopy

Abstract

The viticulture business has an increasing demand for high quality products directly impacting on their market acceptance. Hence, the constant monitoring activities of vineyards, specially quality and maturation stage attributes, are important to assure the production of special wines, justified by their high value-added characteristic. Therefore, optical technologies appear as promising techniques for non-destructive analysis of wine grapes in view of reducing agricultural inputs and analysis duration. The main objective of the present work was to develop predictive models for quality and maturation stage attributes of wine grapes using visible/near infrared (VIS-NIR) reflectance spectroscopy. A total of 432 ‘Syrah’ and 576 ‘Cabernet Sauvignon’ berries were collected and their reflectance spectra were acquired using FieldSpec® 3 spectroradiometer for a spectral range from 450 to 1800 nm. In a posterior step, total soluble solids, total anthocyanins and yellow flavonoids were determined as the reference standards. Before elaborating mathematical models, the spectral data were submitted to several pretreatments, such as smoothing, derivate and corrections. Principal Component Regression (PCR) and Partial Least Squares Regression (PLSR) were utilized as predictive models using both the complete spectrum data (450–1800 nm) and a smaller set of spectral samples selected by the Jack-Knife method. Other predictive models were also developed utilizing Multiple Linear Regression (MLR) with spectral signatures from quality attributes. Maturation stages were discriminated using Principal Component Analysis – Linear Discriminant Analysis (PCA-LDA), Principal Component Analysis – Quadratic Discriminant Analysis (PCA-QDA), Principal Component Analysis – Linear Discriminant Analysis using Mahalanobis distance (PCA-LDA Mahalanobis), and Partial Least Squares Discriminant Analysis (PLS-DA) classification techniques. The construction of the PCR, PLSR and MLR regression models has provided robust predictions for total soluble solids and anthocyanins contents (R2 ≥ 0.90), as well as flavonoids contents with a certain degree of precision (R2 ≥ 0.70). Moreover, it was possible to differentiate distinct maturation stages of vines with 93.15% of accuracy using PLS-DA. Therefore, VIS-NIR reflectance spectroscopy is a powerful tool for non-destructive evaluation of quality and maturation stage attributes of intact grapes from ‘Syrah’ and ‘Cabernet Sauvignon’ grapes.