Fig. 1. The score diagnostic plot of the 110 long chromatograms. The orthogonal distance is plotted vs. the robust distance. The cut-off values are determined in the space of five rPC's.

Fig. 2. One hundred and ten long chromatograms before (a) and after (b) warping (N = 60, t = 3) with peak 1 caffeine, peak 2 (−)-epigallocatechin gallate and peak 3 an unidentified substance, only in high concentrations present in one tea extract.

Fig. 3. One hundred and ten short chromatograms before (a) and after (b) warping (N = 30, t = 3) with peak 1 containing caffeine, peak 2 containing (−)-epigallocatechin gallate and peak 3 an unidentified substance, only in high concentrations present in one tea extract.

Fig. 4. The score diagnostic plot of the 110 short chromatograms. The orthogonal distance is plotted vs. the robust distance. The cut-off values are determined in the space of six rPC's.

Fig. 5. Histogram of the 55 TEAC values.

Fig. 6. Regression coefficients of the PLS model obtained for the long chromatograms.

Fig. 7. Plot of the reference chromatogram for the alignment of the long chromatograms. Only the dark variables are retained in the UVE-PLS model. Peak 1 represents caffeine, peak 2 (−)-epigallocatechin gallate and peak 3 is an unidentified substance, which is only clearly present in chromatograms 61 and 62.

Fig. 8. Results of (a) the PLS model (eight factors) built with the 40 warped long calibration chromatograms, and (b) the UVE-PLS model (seven factors) built with the 40 warped long calibration chromatograms with 142 variables retained (○: calibration samples, *: test samples).

Fig. 9. Regression coefficients of the PLS model obtained for the short chromatograms.

Models constructed for (a) the long, and (b) the short chromatograms and (c) the long chromatograms of which the number of sampling points is reduced by averaging

Fn: model complexity, RMS: root mean squared error, RMSECV: root mean squared error of cross-validation and RMSEP: root mean squared error of prediction.