Merging vibrational spectroscopic data for wine classification according to the geographic origin

  1. Teixeira dos Santos, C.A. 3
  2. Páscoa, R.N.M.J. 3
  3. Sarraguça, M.C. 3
  4. Porto, P.A.L.S. 1
  5. Cerdeira, A.L. 1
  6. González-Sáiz, J.M. 2
  7. Pizarro, C. 2
  8. Lopes, J.A. 4
  1. 1 CVRVV, Comissão de Viticultura da Região dos Vinhos Verdes, Rua da Restauração, 318, Porto, Portugal
  2. 2 Universidad de La Rioja
    info

    Universidad de La Rioja

    Logroño, España

    ROR https://ror.org/0553yr311

  3. 3 Universidade Do Porto
    info

    Universidade Do Porto

    Oporto, Portugal

    ROR https://ror.org/043pwc612

  4. 4 Universidade de Lisboa
    info

    Universidade de Lisboa

    Lisboa, Portugal

    ROR https://ror.org/01c27hj86

Mostrar afiliaciones +
Revista:
Food Research International

ISSN: 0963-9969

Año de publicación: 2017

Volumen: 102

Páginas: 504-510

Tipo: Artículo

DOI: 10.1016/J.FOODRES.2017.09.018 SCOPUS: 2-s2.0-85029652502 WoS: WOS:000419418100059 GOOGLE SCHOLAR

Otras publicaciones en: Food Research International

Resumen

The wine making procedure is no longer a secret and it is nowadays well described and repeated around the world. Nevertheless, wines present unique features, strongly associated with their geographic origin. Classification systems were developed to catalogue wines according to the provenance, and are currently established by official authorities in order to ensure wine authenticity. The use of near-infrared (NIR), mid-infrared (MIR) and Raman spectroscopy for tracing the origin of wine samples, has been reported with different levels of success. This work evaluated and compared the performance of these techniques, as well as their joint use, in terms of geographic origin classification. NIR, MIR and Raman spectra of wine samples belonging to four Portuguese wine regions (Vinhos Verdes, Lisboa, Açores and Távora-Varosa) were analyzed by partial least squares discriminant analysis (PLS-DA). Results revealed the better suitability of MIR spectroscopy (87.7% of correct predictions) over NIR (60.4%) and Raman (60.8%). The joint use of spectral sets did not improve the predictive ability of the models. The best models were achieved by combining MIR and NIR spectra resulting in 86.7% of correct predictions. Multiblock partial least squares (MB-PLS) models were developed to further explore the combination of spectral data. Although these models did not improve the percentage of correct predictions, they demonstrated the higher contribution of MIR spectroscopic data, in the development of the models. © 2017 Elsevier Ltd