Eve Laroche-Pinel, Erica Sawyer, Benjamin Corrales, Kaylah Vasquez, Monica Cooper, Marc Fuchs, and Luca Brillante*
*Department of Viticulture and Enology, California State University Fresno, 2360 E Barstow Ave, Fresno, CA, 93740 (lucabrillante@csufresno.edu)

Grapevine viruses are impacting vineyards, affecting fruit ripening, decreasing grape quality, and reducing yield. Identifying infected vines is crucial to limit virus spread. Remote sensing can measure the biophysical and biological properties of vegetation that are affected by viruses. We evaluated the potential of hyperspectral VIS/NIR imagery to detect red blotch-infected vines. We tagged and geolocated hundreds of vines in a vineyard in Napa, California. For each vine, we sampled leaves in 2020 and 2021 to identify infection using PCR analysis. At the same time, for the same vines, we took pictures of the canopy sides with a VIS/NIR hyperspectral camera (from 500 to 900 nm) mounted on a tripod (n ~ 700) and above the canopy with a drone. To classify infection in vine images, we tested different machine-learning algorithms, partial least square discriminant analysis (PLS-DA), random forest (RF), support vector machine (SVM), and their ensembles, obtained by simple averaging of predictions and stacks using PLS-DA as the input model. For images taken with the tripod, the best accuracies are reached using the PLS-SVM ensemble. For the entire data set, the accuracy reaches 69.5%, with 80% accuracy for the non-infected class. For the late season, the accuracy reaches 74.5%, with an accuracy of 83% for the non-infected class. Feature importance computation shows that the bands located in the red, green, and infrared domains are most informative for the model. The findings of this project are encouraging, as they have the potential to introduce a new method for inspecting vineyards, allowing detection of virus transmission and identification of suspect vines (predicted to be infected). Where a more cautious approach is necessary before removal, PCR analysis can be conducted on the targeted vines.

Funding Support: CDFA SCBGP; California State University - Agricultural Research Institute