Craig Ebersole,* Sonet Van Zyl, Stephan Sommer, and Qun Sun *California State University, Fresno, 2360 E. Barstow Avenue, M/S VR89, Fresno, CA, 93740 (cynthiaw@csufresno.edu)

For wineries processing hand-harvested grapes, a visual inspection of microbial grape spoilage is expedient and cost-effective. However, increased adoption of machine harvesters, which frequently rupture berry skins and make visual inspection less precise, and the high degree of error of visual inspections between individuals make a quantitative approach to assess spoilage necessary. Fourier transform infrared spec­troscopy, combined with multivariate analysis, is being investigated as an approach to predict grape health, as a sample can be analyzed in less than one minute. Calibration data was obtained from grape samples of Chardonnay, Riesling, Petite Sirah, and Zin­fandel that were sorted into fractions of 0, 5, 10, 15, or 20% microbially-impacted clus­ters in healthy grape material and run on the FOSS Winescan. Spectral data was ana­lyzed in R Studio using the partial least squares package with spoilage level being the dependent variable. Attributes under consideration include volatile acidity, gluconic acid, ethanol, lactic acid, glucose-fructose content, total soluble solids, titratable acidity, tartaric acid, malic acid, pH, and yeast assimilable nitrogen. Based on previous research, positive correlations have been found between the degree of microbial spoilage and volatile acidity, gluconic acid, ethanol, lactic acid, and glucose-fructose content, so these predicting factors will be overweighted. A model will be selected that optimizes for a high correlation coefficient and a low difference between the root mean squared error of the model and the ten-fold cross-validated model. As the correlation improves, by using more predicting factors with training data, the pre­dictive power of the model on new data decreases, due to low correlation between some predicting factors and the dependent variable. Therefore, the ideal combination of predicting factors, predicting factor weights, and number of components will be selected to provide accurate Grape Health Index scores.

Funding Support: California Winegrape Inspection Advisory Board & The Agricultural Research Institute