Spectral differences of grapevine leaves and clusters affect prediction accuracy of cluster architecture and juice traits

Abstract

Grapevine productivity is increasingly threatened due to climate change. Grapevine clusters are especially sensitive to fungal infections such as Botrytis bunch rot and to abiotic stresses including sunburn damage. In this study, we evaluated how hyperspectral reflectance from different grapevine organs contributes to the prediction of cluster architecture and juice quality traits to support high-throughput phenotyping in Riesling and Pinot clonal populations. Using partial least squares regression (PLSR), we assessed the prediction accuracy of eight cluster architecture traits and six juice traits under two data analysis strategies: data splitting by population and by cluster type. Both sets of traits exhibited strong population-specific trends, as prediction accuracy depended on the data split strategy and the organ reflectance measurements used to train the model. Cluster reflectance models consistently outperformed those trained on dry leaf reflectance for most traits, except pH. Cluster type data split increased variance and improved calibration results for several traits, including number of berries (R2 = 0.53), berry diameter (R2 = 0.79) and total acidity (R2 = 0.48), whereas population-based splits resulted in lower prediction accuracy. Variable importance in projection (VIP) scores indicated that visible (400-700 nm) and red-edge (680-760 nm) spectral regions contributed most to cluster architecture predictions, while near- and shortwave infrared spectra (700-1200 nm) were more relevant for juice trait prediction. Overall, our results demonstrate that organ-specific hyperspectral reflectance combined with appropriate data analysis strategies can enable scalable, non-destructive screening of cluster architecture and fruit quality traits to support early selection decisions for more efficient allocation of labour and costs in grapevine breeding programs.