The objective of this work was to compare the effectiveness of three chemical families, namely, chlorogenic acids, fatty acids, and elements, for the discrimination of Arabica varieties (traditional versus modern introgressed lines) and potential terroirs within a given coffee-growing area. The experimental design included three Colombian locations in full combination with five (one traditional and four introgressed) Arabica varieties and two field replications. Chlorogenic acids, fatty acids, and elements were analyzed in coffee bean samples by HPLC, GC, and ICP-AES, respectively. Principal component analysis and discriminant analysis were carried out to compare the three methods. Although elements provided an excellent classification of the three locations studied, this chemical class was useless for Arabica variety discrimination. Chlorogenic acids gave satisfactory results, but fatty acids clearly offered the best results for the determination of both varieties and environments, with very high percentages of correct classification (79 and 90%, respectively).

In a previous study, the effectiveness of chlorogenic acids, fatty acids (FA), and elements was compared for the discrimination of Arabica varieties and growing terroirs. Since FA provided the best results, the aim of the present work was to validate their discrimination ability using an extended experimental design, including twice the number of location x variety combinations and 2 years of study. It also aimed at understanding how the environment influences FA composition through correlation analysis using different climatic parameters. Percentages of correct classification of known samples remained very high, independent of the classification criterion. However, cross-validation tests across years indicated that prediction of unknown locations was less efficient than that of unknown genotypes. Environmental temperature during the development of coffee beans had a dramatic influence on their FA composition. Analysis of climate patterns over years enabled us to understand the efficient location discrimination within a single year but only moderate efficiency across years.