Optimization of FTIR-ATR spectroscopy for botanical authentication of unifloral honey types and melissopalynological data prediction

• Published in: European food research & technology 2015-06, Vol.240 (6), p.1101-1115
• Main Authors: Svečnjak, Lidija, Bubalo, Dragan, Baranović, Goran, Novosel, Hrvoje
• Format: Article
• Language: English
• Subjects: Agriculture; Analysis; Analytical Chemistry; Bees; Biodiversity; Biotechnology; Botany; Chemistry; Chemistry and Materials Science; Data processing; Discriminant analysis; Flowers & plants; Food; Food Science; Forestry; Fourier transforms; Honey; Identification; Investigations; Methods; Optimization; Original Paper; Pollen; Principal components analysis; Spectroscopy; Statistical analysis; Studies
• ISSN: 1438-2377; 1438-2385
• DOI: 10.1007/s00217-015-2414-1

Six hundred thirty-two honey samples from Croatia were analyzed by reference methods (physicochemical and melissopalynological analyses), from which 506 were assigned to one of nine unifloral honey types (black locust, sweet chestnut, lime, sage, heath, mandarin, false indigo, immortelle and winter savory). Vibrational spectra of honey samples recorded using Fourier transform infrared (FTIR) spectroscopy and attenuated total reflection (ATR) technique were subjected to comprehensive chemometric assessment involving optimization of data preprocessing and multivariate statistical analyses, namely principal component analysis, partial least squares regression, partial least squares-discriminant analysis and receiver operating characteristic (ROC) analysis. A specially designed hybrid multidimensional model (partial least squares-linear discriminnat analysis) was used to assess a relationship between honey IR spectra and relative frequencies of pollen grains (pollen spectrum) present in selected honey samples. The study demonstrated that a targeted spectral profiling, i.e., preprocessing of a fingerprint spectral region (1,800–700 cm −1 ) by particular Savitzky–Golay filtering parameters and ROC modelling approach, enables accurate classification (100 % correct classification) of unifloral honeys by their botanical origin with low misclassification risk (MR = 0.099) and negligible small number of misclassified samples (1–2 per class—honey type). The results have also revealed good predictive strength of FTIR-ATR spectroscopy for melissopalynologycal data.