Identifying the boundaries of the sensory space of red Bordeaux wines using an innovative machine learning approach. Application to the identification of new varieties adapted to climate change.

Abstract

Climate change is likely to impact wine typicity across the globe, raising concerns in wine regions historically renowned for the quality of their terroir. Among potential changes, modifying plant material (i.e., clones, rootstocks and varieties) is considered to be one of the most promising potential levers for adaption to climate change. However, changing varieties raises the issue of how to protect the region’s wine typicity. In Bordeaux (France), extensive studies focus on the identification of varieties that would be more suitable for a warmer and drier climate, but less research has been done concerning their impacts on Bordeaux wine typicity. The present study, conducted on 26 red varieties - including traditional-Bordeaux varieties and non-Bordeaux varieties - over five vintages, aimed to characterise Bordeaux wine typicity and to identify non-Bordeaux varieties that could fit into this sensorial space. Through a random forest analysis of the results of a Check-All-That-Apply (CATA) sensorial analysis and a typicity profiling with an exceptionally large panel of experts (approximately 40 judges), the typical red Bordeaux sensory space was clearly defined with consensus among the judges. Interestingly, one non-Bordeaux variety was found to produce wines with high Bordeaux typicity. Finally, via hierarchical clustering analyses based on multiple correspondence analysis, five non-Bordeaux varieties that produce wines with similar sensorial spaces to the traditional Bordeaux varieties were identified. These results indicate that these cultivars could be introduced to the Bordeaux grape-mix without profoundly altering Bordeaux wine typicity in a context of climate change, if found to be suitable for a warmer climate than the current climate of the Bordeaux region.

Keywords

• Climate change
• Wine typicity
• Sensory analyses
• Terroir
• Random forest
• CATA