Infrared spectroscopy investigation of fresh grapevine organs for clustering and classification.

Abstract

The spectral information acquired from fresh whole grapevine organs have yet to be fully explored. Infrared spectroscopy provides the means to rapidly measure fresh plant material and providing extensive information on the physical and chemical structure of samples. This study aimed to explore the spectra of fresh grapevine shoots, leaves, and berries throughout the growing season for clustering and classification. Sampling was performed across two vintages (2019-2020; 2020-2021) from November to March. Five locations, seven cultivars, and 17 commercial vineyards were included. Collection of whole shoots, including leaves and grape bunches, were performed on a monthly basis. The fresh grapevine organs were analysed using three spectroscopy methods within 24-36 hours of sampling. Mid-infrared (MIR) and near-infrared (NIR), making use of a solid probe (NIR-SP) and a rotating sphere (NIR-RS), were investigated. The raw spectra were firstly investigated using principal component analysis (PCA) followed by a more novel chemometric approach, unsupervised
self-organising maps (SOM). PCA as well as unsupervised SOM showed the most considerable grouping based on organ type. Additionally, the unsupervised SOM showed separation trends based on phenological stage. Investigation of the datasets per organ with SOM showed separation based on the phenological stage for berries and shoots, as well as shoots clustering based on lignification. Supervised SOM were examined for classification based on the observed clustering per organ type, phenological stage, and lignification. The accurate prediction of organ at 90.3% was possible for the NIR-SP dataset for 2019-2021. Overlapping of various phenological stages were seen for the grape berry datasets, but prediction improved to 85.6% for the NIR-RS 2019-2021 dataset when certain phenological
stages were grouped together. Accurate predictions of lignified and unlignified shoots were also seen for the NIR-SP 2019-2021 and NIR-RS 2020-2021 datasets at 74.4% and 89.9% respectively. The possibility of using spectral variable selection to improve the supervised SOM predictions were explored and promising results obtained for certain datasets. Following variable selection with OPLS-DA and S-plots, the prediction of shoots and leaves improved by 14% for the NIR-RS 2020-2021 dataset. The prediction of lignified and unlignified shoots improved considerably to 92.3% for the NIR-SP 2019-2021 dataset and 95.9% for the NIR-RS 2020-2021 dataset. This study showed the extensive information available in infrared spectra of fresh grapevine organs and how the information could be used to achieve important clustering and classifications objectives