Hyperspectral imaging and machine learning for monitoring grapevine physiology

To face emerging competition and challenges, wine producers globally rely on precision viticulture (PV) solutions to boost productivity, enhance quality, increase profitability, and reduce the environmental impact of vineyards. Current pv methods predominantly use multispectral sensor data from several platforms (satellites or vineyard installations). However, these applications generally use data analysis strategies lacking physiological grapevine support.

Current remote sensing strategies rely heavily on reflectance data and energy balance modelling using thermal imagery to estimate crop water use and stress. These approaches show great promise for driving precision management decisions, but still require work to better understand how detected changes relate to meaningful physiological changes. Under water stress, grapevines exhibit a range of responses involving both biological and physical changes within leaves and canopies.

In the context of climate change, it is essential to provide producers with alternatives based on local grape varieties capable of meeting modern quality and sustainability requirements.

The study of wine evolution during barrel aging is an important aspect of wine quality. Our previous works have shown that wine metabolome monitoring by

1H-NMR approaches allows determining the impact of different winemaking processes including traitements using enzymes or finning agents [1].

In this study we have compared the predictive ability of two phenological models: a traditional Thermal Time (TT) and a version of the more recently develop Unified Model (UM).