Inter-correlation of apple firmness determinations and development of cross-validated regression models for prediction of sensory attributes from instrumental and compositional analyses.

The texture of apples is paramount for determining fruit quality. This research explored the correlations among firmness determinations from the Sinclair iQ™ Firmness Tester (SiQ™), the Aweta Acoustic Firmness Sensor (AFS), and eight measurements from the Mohr Digi-Test-2 (MDT) instrument. Assessments were conducted on a collection of nine apple cultivars (Ambrosia, Aurora Golden Gala™, Honeycrisp, Fuji, Imperial Gala, McIntosh, Pink Lady™, Silken, Salish™), with a broad range of firmness values, in each of two years. Sensory analysis of the apples was conducted using a semi-trained panel (n = 10) to evaluate crispness, hardness, juiciness and skin toughness, in quadruplicate at two testing dates, providing eight data points per cultivar per year. Inter-correlations of the instrumental firmness determinations (SiQ™, AFS, MDT) revealed that most values were highly correlated with one another (r > 0.500 n = 72). This suggested that the instruments were tracking similar, but not identical, underlying characteristics. Multiple regression models were developed using the 2016 data to predict the sensory attributes from the instrumental and compositional (titratable acidity, soluble solids concentration, absorbed juice) analyses. Models with the highest R were cross-validated using the 2015 data. Accuracy of these models was evaluated using R and prediction standard errors (PSEs) - an index quantifying the difference between the predicted and actual values. In general, simple 1- and 2-variable models satisfactorily predicted hardness and crispness, with the R values ranging between 85 and 89%, while more complex non-linear models were required to predict juiciness and skin toughness. Correlations coefficients reported in this research allow for interconversion of experimental firmness data, as determined by the SiQ™, AFS and MDT. Regression models predicting hardness, crispness and juiciness from instrumental/compositional analyses, revealed that the quality factor (QF) variable was particularly important for estimation of textural characteristics. Therefore the MDT, among the instruments evaluated, was the instrument of choice for quality assessment of apples. Since cross-validation of the models accounted for a high proportion of the variance (70-82%) in a new data set with small PSEs (2.67-6.36) (on a 100-unit scale), the developed models were appropriate for estimating the apple textural attributes.