Delineation of coffee terroir zones using machine learning and support from portable X-ray fluorescence

• Grant number: 25/26213-8
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: December 01, 2025
• End date: November 30, 2026
• Field of knowledge: Agronomical Sciences - Agronomy - Soil Science
• Principal Investigator: Michele Duarte de Menezes
• Grantee: Pedro Tadeu Leite
• Host Institution: Escola de Ciências Agrárias. Universidade Federal de Lavras (UFLA). Lavras , SP, Brazil
• Associated research grant: 21/06968-3 - From seed to cup: internet of things technology in the quality coffee production chain, AP.TEM

Abstract

The term terroir, widely disseminated in the world of wines and also applicable to coffees, reflects the conditions that affect the plant, such as climate, soil, relief, temperature, and humidity - as well as human factors, such as agricultural production practices and post-harvest processing. When these factors interact with plant genetics, unique characteristics are generated, which are reflected in the flavor and aroma of the coffee. Given the multivariate and complex nature of the interactions among factors that influence beverage quality in the definition of terroir, the use of machine learning techniques shows great potential when applied to unravel, quantify, and understand data in operational agribusiness environments (Patrício and Rieder, 2018). Moreover, these techniques have the potential to convert sensor data obtained rapidly, more cheaply, and in a non-polluting manner into useful or desirable information. This forms the theoretical framework of the so-called soft sensors, a term that emerged from the combination of the words software-since sensor signal evaluation models are generally implemented in computer programs-and sensor, because these models generate information similar to that produced by hardware sensors. The portable X-ray fluorescence soft sensor (pXRF) is capable of detecting the total contents of several chemical elements of the periodic table (from Mg to U), as each element has a characteristic fluorescence energy. Data are obtained rapidly (approximately 1 minute) through proximal detection with no or minimal sample pre-processing (Silva et al., 2021), which constitutes the main advantage of this sensor. In the context of coffee cultivation, special attention will be given to the pXRF contents of K, Cl, and S, since these chemical elements are important from the standpoint of beverage quality. Potassium (K) has long been considered the element of quality in plant nutrition (Malavolta, 1980). Amorim et al. (1973) reported that in coffee plants fertilized with potassium chloride, the Cl¿ anion may exert a negative effect on beverage quality. Sulfur (S), in turn, is related to caffeine contents. (AU)