Assessing bunch architecture for grapevine yield forecasting by image analysis 

Abstract

Context and purpose of the study – It is fundamental for wineries to know the potential yield of their vineyards as soon as possible for future planning of winery logistics. As such, non-invasive image-based methods are being investigated for early yield prediction. Many of these techniques have limitations that make it difficult to implement for practical use commercially. The aim of this study was to assess whether yield can be estimated using images taken in-field with a smartphone at different phenological stages. The accuracy of the method for predicting bunch weight at different phenological stages was assessed for seven different varieties.

Material and methods – During the 2017-18 growing season in the Coombe Vineyard at the Waite Campus of the University of Adelaide seven different varieties were chosen for this study: Semillon, Grenache, Shiraz, Merlot, Sauvignon Blanc, Tempranillo and Cabernet Franc. After fruitset, 30 vines per variety were selected and two shoots were flagged on each vine. Images of bunches were taken five times from EL stage 30-31 to EL stage 37-38 using a smartphone. Bunch volumes were estimated from images. At harvest bunches were collected, weighed and imaged in the laboratory to compare with field images.

Results – This new approach using a smartphone to forecast the yield showed promising results. Accurate weight forecast models could be obtained by taking bunch images at veraison (R2 ranging from 0.71 to 0.84). As the bunch architecture of different varieties can vary further studies are required to improve the accuracy of this method. The tools used for this study are inexpensive, in common use, and do not need a high level of expertise to use them, furthermore, the labour required to obtain data, is not time-consuming.