Table 1. A statistical summary of discriminant models computed to classify viable, empty and petrified seeds of L. sibirica into three and two classes. | |||||

Class | A | R^{2}X_{p} | R^{2}X_{0} | R^{2}Y | Q^{2 }_{cv} |

Three | 2 + 13 | 0.267 | 0.733 | 0.842 | 0.820 |

Two | 1 + 12 | 0.041 | 0.959 | 0.937 | 0.931 |

A = number of significant components to build the model (the first and second values represent the number of components for predictive and orthogonal variations, respectively); R^{2}X_{p} = the predictive spectral variation; R^{2}X_{o} = Y-orthogonal variation (spectral variation uncorrelated to class discrimination) for the OPLS-DA models; R^{2}Y = the modelled class variation; and Q^{2}_{cv} = the predictive power of a model according to cross validation. |

Table 2. Predicted class membership of viable, empty and petrified seeds in the test sets by three-class OPLS-DA modelling of single seed NIR spectra. | ||||||||||

Seed lot fractions | No. of test | 1 | 2 | 3 | No class | 1 & 2 | 1 & 3 | 2 & 3 | 1 & 2 & 3 | Classification accuracy (%) |

Viable (1) | 45 | 44 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 98 |

Empty (2) | 45 | 0 | 37 | 5 | 1 | 0 | 0 | 2 | 0 | 82 |

Petrified (3) | 45 | 0 | 1 | 39 | 4 | 0 | 0 | 1 | 0 | 87 |