**5.3 Statistical analysis**

The hyperspectral reflectance and fluorescence data of the control and treated plants were subjected to statistical analysis through the Student's t-criterion and linear stepwise Discriminant Analysis (DA). Because 1170 reflectance and 910 fluorescence values were available to be used as classification features, it was computationally efficient to select a subset of bands on the basis of discriminant capability.

The reflectance analysis was performed in four most informative for investigated plants spectral ranges: green (520-580 nm, maximal reflectivity of green vegetation), red (640-680 nm, maximal chlorophyll absorption), red edge (680-720 nm, maximal slope of the reflectance spectra) and the NIR (720-770 nm) (Krezhova et al., 2005, 2007). The statistical significance of the differences between SRC of control and treated plants was examined in eight spectral bands (wavelengths) chosen to be disposed uniformly over the above mentioned ranges (λ1 = 524.29 nm, λ2 = 539.65 nm, λ3 = 552.82 nm, λ4 = 667.33 nm, λ5 = 703.56 nm, λ6 = 719.31 nm, λ7 = 724.31 nm, and λ8 = 758.39 nm).

The fluorescence spectra were analyzed in five characteristic spectral bands, chosen at wavelengths: 1 (at the middle of the forefront edge), 2 (first maximum), 3 (at the middle between first and second maximum), 4 (second maximum), and 5 (at the middle of the rear slope). They are illustrated in Fig. 11 for a typical fluorescence spectrum of green vegetation.

The Student's t-criterion and linear DA were applied for determination of the statistically significance of differences at p<0.05 between the means of sets of the values of the reflectance and chlorophyll fluorescence of control and treated plants in the above mentioned wavelengths. They were further regarded as discriminative features. The Student's t-criterion was utilized under the prerequisite for the existence of numerous, independent and approximately of one and the same order factors of small impacts on the variables under examination. DA was used to increase classification accuracy. One output of the method is the determination of the posterior probability that spectral data of a given leaf belongs to the class of control or treated plants. For this purpose discriminant analysis will be implemented in one dimensional spaces defined by the features examined. In some of the cases DA was performed in two or three-dimensional spaces for enhancement of the discriminative possibility.

Fig. 11. Characteristic wavelengths chosen for statistical analysis of fluorescence.

All statistical analyses were conducted using the STATISTICA software, Version 6.1, 2002, (http://www.statsoft.com).
