Decoding local terroir: Data mining to predict sensory profiles of coffee beverage

Abstract

CONTEXT

Secondary metabolism aids plant adaptation to environmental stress, suggesting that terroir may reflect stress-induced coffee crop responses. Additionally, soil and environmental conditions affect plant growth, which in turn influences bean composition and beverage typicity.

OBJECTIVE

To explore data mining to promote concise pattern discovery and assist in the farm routine driven by coffee beverage typicity in a commercial farm, in the Campos das Vertentes indication of origin, Minas Gerais state, Southeastern Brazil.

METHODS

Firstly, a multivariate dataset was created for two harvest seasons (2019/2020 and 2020/2021) to characterize seven local terroir zones. The dataset included information of soil parent material (via airborne gamma-ray spectroscopy), soil properties (chemistry, physics, and magnetic susceptibility via proximal sensing), relief (via digital terrain models), leaf data (ICP-OES and NDVI via remote sensing) coupled with coffee beverage analysis (sensory and chemical analysis) of Coffea arabica (cultivar Acaiá IAC 474–19) in a pilot farm. Data mining tools were explored to identify concise pattern discovery and assist in the farm routine driven by coffee beverage typicity. Thus, MFA was used to assess whether beverage sensory profiles could distinguish terroir zones at the local scale. Random forest-RFE identified key soil-environmental attributes linked to sensory notes.

RESULTS AND CONCLUSION

Three terroir zones were clustered based on beverage intensity and dominant sensory notes. Random forest accurately predicted fruity and caramel notes. The B content in the soil subsurface was ranked for the fruity note, while magnetic susceptibility, organic matter, and the topographic wetness index were ranked for the caramel. A rough undesirable note was linked to P, Mg, and Zn content in the soil subsurface. Overall, soil fertility at the subsurface and relief primarily drive coffee typicity and should guide farm precision management.

SIGNIFICANCE

This study demonstrates that data mining techniques can enhance precision agriculture by stratifying agricultural systems and facilitating terroir characterization. This approach enables site-specific management, improves coffee quality, and supports certification processes.