Linear Discriminant Analysis of spectral measurements for discrimination between healthy and diseased trees of <i>Olea europaea</i> L. artificially infected by <i>Fomitiporia mediterranea</i>
Anhelina Zapolska, Chariton Kalaitzidis, Emmanouil A. Markakis, Eleftherios K. Ligoxigakis, Georgios Koubouris
Abstract
Fomitiporia mediterranea , commonly known as ‘Esca’, is a detrimental fungus for many tree species and grapevine and is considered to be one of the main causal agents of wood decay of olive plantations in the Mediterranean region. Symptomatic trees are mainly identified at the advanced stages of the disease, so no curative control measures can be applied. In recent years, spectral measurements have been used in agriculture for early detection of disease incidents. In this paper, Linear Discriminant Analysis (LDA) of hyperspectral data collected in situ by a ASD FieldSpec® 3 spectroradiometer was used to investigate its potential for identifying alterations caused by the fungus in the most popular Greek olive cultivars ‘Amfissis’, ‘Chalkidikis’, ‘Mastoidis’, ‘Koroneiki’ and ‘Kalamon’. In order to identify the optimal wavelength ranges for LDA that are indicative of Fomitiporia mediterranea presence, Principal Component Regression (PCR) and Partial Least Square Regression (PLSR) were applied. The results showed a good classification accuracy of infected and non-infected trees. Furthermore, trees that were not identified as diseased through laboratory analysis (reisolation from trunks) but had been artificially infected at the beginning of the experiment, were actually classified as infected by LDA, implying that hyperspectral scanning was able to identify past infection even if the pathogen was no longer present.