Visible-near infrared spectroscopy based Grapevine leafroll-associated virus-3 detection from undetached leaves under field condition

Abstract.

Grapevine leafroll disease (GLD) causes substantial economic losses to grapevine industry in the United States. Methods like destructive sampling followed by enzymelinked immunosorbent assay (ELISA), reverse transcriptionpolymerase chain reaction (RTPCR), and destructive spectral measurements have been used to detect GLD. Applicability of visible and near infrared (VISNIR) spectrometer was assessed to detect GLD in red berried winegrape cultivar, Cabernet Sauvignon. On each sampling day, 60 healthy and 60 infected leaves (30 symptomatic and 30 presymptomatic) were sampled using VISNIR spectrometer. Feature extraction using stepwise multilinear regression (SMLR) and partial least square regression (PLSR) reported significant differences between the healthy and infected plants in visible (351, 377, 501, 526, 626, and 676 nm) and near infrared (726, 826, 901, 951, 976, 1001, 1027, 1052 and 1101 nm) regions. However, significant spectral bands from near infrared region (901, 1001, 1027 and 1101 nm) repeated temporally for both symptomatic and presymptomatic samples. Principal component analysis (PCA) suggested that selected spectral bands exhibit high class discriminating power for disease detection. Classification accuracies in the range of 80–100%, 63–100% and 77–100% was reported for healthy, infected and overall classes using quadratic discriminant analysis (QDA) and naïve bayes (NB). The accuracy for disease detection was significantly higher during early stages compared to later stage, indicating robustness of the selected features for early disease detection. This results from this study could be helpful in developing a portable and low cost sensing system for rapid and non–destructive detection of GLD infestation in redberried grapevines.