Tavares L. A., Ferreira A. G., Ferreira M. M. C., Correa A, Mattoso L. H., "Analysis of blends of raw coffees of Arabica and Robusta varieties through ¹H NMR and chemometric methods". Copenhagen, Denmark, 13-15/09/2004: 7th International Conference on Applications of Magnetic Resonance in Food Science: The multivariate challenge, Poster 24. Proceedings in book: S.B. Engelsen, P.S. Belton, H.J. Jakobsen (Eds.): Magnetic Resonance in Food Science: The Multivariate Challenge, Special Publication No. 299, The Royal Society of Chemistry , London, pp. 80-88 (2005). Awarded Poster 24.
[Article. More information about the meeting see at Notes2.]

24 - BLENDS ANALYSIS OF ARABICA AND ROBUSTA RAW COFFEES SPECIES THROUGH ¹H NMR AND CHEMOMETRIC METHODS

Leila Aley Tavares 1, Elisangela Fabiana Boffo 1, Antonio Gilberto Ferreira 1, Márcia Miguel Castro Ferreira 2,
Luiz Henrique Capparelli Mattoso 3, Alessandra Alves Corrêa 3
1 UNIVERSIDADE FEDERAL DE SÃO CARLOS, 2 UNIVERSIDADE DE CAMPINAS, 3 EMBRAPA, BRAZIL

The drink coffee is prepared from arabica or robusta variety grain, or even from blends of them. Arabica variety presents flavor more pronounced and refined, so it can get a best price than robusta in the international trade. This increase the chance of frauds and require techniques that have power to discriminate both varieties in a blend.
In this work was used chemometric methods and 1H NMR spectroscopy data to discriminate arabica and robusta blends as much as classificatory and quantitative analysis.
The ¹H NMR data were aquire in BRUKER 9.4 Tesla equipment and all measurements were done in triplicate. Were used eleven coffee samples without roast and with different contents of arabica and robusta specie. The chemometric methods used were: PCA and HCA for exploratory analysis, KNN and SIMCA for classificatory analysis, PCR and PLS for quantitative.
Results obtained showed that the methods of exploratory and classificatory analysis can group the blend samples correctly in two groups, one content less and the other high proportion of arabica specie. However the method KNN shows be better than SIMCA because it can discriminate the samples in two groups (arabica and robusta) without any error using ten neighbours. To predict the proportion of each variety in the coffee blend were used PLS and PCR methods and PLS showed better than the PCR . The same were done with roast coffee sample but the results were not so good because the roast degree made a strong influence in the chemometric analysis.