Prediction models for sunscald on apples (Malus domestica Borkh.) cv. Granny Smith using Vis-NIR reflectance

Abstract

Abstract Sunscald on apples (Malus domestica Borkh) is a physiological disorder that develops post-harvest in susceptible cultivars such as Granny Smith. Symptoms appear on the skin as dark brown areas over fruit sections that were sun exposed during its growth on the tree. Currently, there is no chemical control to prevent its appearance. The objective of this study was to develop a prediction model for this disorder using Vis/NIR reflectance. During 2013/14 and 2014/15 Granny Smith apples were harvest and classified according to their sun exposure in the field (shaded, sun-exposed) and sun injury symptom (mild, moderate, severe), and stored at 0–1\u2009°C (> 90% RH) for 120 d. Every 15 d sunscald incidence was recorded, and spectral fingerprint taken using an optical portable spectrometer Vis/NIR (350–1100\u2009nm). Multivariate analysis was performed using partial least squares discriminant analysis (PLS-R) and intervals PLS-DA (iPLS-DA) in order to develop calibration models and applied them to a prediction set. Two calibration models were developed. The first one (A; R2CV\u2009=\u20090.41) for fruit at harvest with no sun damage symptoms (shaded and sun-exposed, n\u2009=\u2009161), which was able to predict the appearance of sunscald 45 d in advance. The spectral ranges responsible for discrimination (iPLS-DA) of fruit with (Class 1) and without (Class 0) sunscald postharvest were: 500–525\u2009nm, 550–575\u2009nm, 650–675\u2009nm, 775–800\u2009nm, 850–875\u2009nm, and 925–950\u2009nm. The second model (B; R2CV\u2009=\u20090.59) was developed for mild sun damaged fruit (n\u2009=\u2009100), which predicted sunscald 30 d in advance. In this case, the spectral ranges responsible of discrimination were 500–525\u2009nm, 650–675\u2009nm, and 775–800\u2009nm, related to pigments and structural properties of the tissue. The RMSEP for the independent fruit set validation models (n\u2009=\u200969) were 3-4-times higher than that of the calibration models, probably because biological variability. A greater number of independent fruit sets will be required to increase models’ robustness and prediction capability for the industry adoption.