Detection and Quantification of Grapevine Bunch Rot Using Functional Data Analysis and Canonical Variate Analysis Biplots of Infrared Spectral Data

Grapevine bunch rot assessment has economic significance to wineries. Industrial working conditions require rapid assessment methods to meet the time constraints typically associated with grape intake at large wineries. Naturally rot-affected and healthy white wine grape bunches were collected over five vintages (2013 to 2016, 2020). Spectral data of 382 grape must samples were acquired using three different, but same-type attenuated total reflection mid-infrared (ATR-MIR) ALPHA spectrometers. The practical industrial problem of wavenumber shifts collected with different spectrometers was overcome by applying functional data analysis (FDA). FDA improved the data quality and boosted data mining efforts in the sample set. Canonical variate analysis (CVA) biplots were employed to visualise the detection and quantification of rot. When adding 90 % alpha-bags to CVA biplots minimal overlap between rot-affected (Yes) and healthy (No) samples was observed. Several bands were observed in the region 1734 cm(-1) to 1722 cm(-1) which correlated with the separation between rot-affected and healthy grape musts. These bands connect to the C=O stretching of the functional groups of carboxylic acids. In addition, wavenumber 1041 cm(-1), presenting the functional group of ethanol, contributed to the separation between categories (severity % range). ATR-MIR could provide a sustainable alternative for rapid and automated rot assessment. However, qualitative severity quantification of rot was limited to only discriminating between healthy and severe rot (>= 40 %). This study is novel in applying FDA to correct wavenumber shifts in ATR-MIR spectral data. Furthermore, visualisation of the viticultural data set using CVA biplots is a novel application of this technique.