Discriminating Xylella fastidiosa from Verticillium dahliae infections in olive trees using thermal- and hyperspectral-based plant traits

Globally, olive (Olea europaea L.) productivity is threatened by plant pathogens, particularly the fungus Verticillium dahliae (Vd) and the bacterium Xylella fastidiosa (Xf). Infections by these pathogens restrict water and nutrient flow through xylem, producing a similar set of symptoms that can also be confounded with water stress. Conventional in situ monitoring techniques are time consuming and expensive, necessitating the development of large-scale detection methods. Airborne hyperspectral and thermal imagery have been successfully used to detect both Xf and Vd infection symptoms independently, i.e., when only one of the two diseases is present. Nevertheless, the discrimination of Vd from Xf infections in contexts where both pathogens are present has not been addressed to date. This study proposes a three-stage machine learning algorithm to distinguish Vd infections from Xf infections, using a series of datasets from 27 olive orchards affected by Xf and Vd outbreaks in Italy and Spain between 2011 and 2017. Plant traits were derived from airborne hyperspectral and thermal imagery, including physiological indices from radiative transfer model inversion, Solar-induced Fluorescence emission (SIF@760), the Crop Water Stress Index (CWSI), and a selection of narrow-band hyperspectral indices. Several distinct spectral traits successfully discriminated Xf from Vd infections. The three-stage method generated a false-positive rate of 9%, an overall accuracy (OA) of 98%, and a kappa coefficient (kappa) of 0.7 when identifying Vd infections using a mixed Vd + Xf dataset. When identifying Xf infections, the false-positive rate was 4%, the OA was 92%, and kappa was 0.8. These results indicate that hyperspectral and thermal traits can be used to discriminate Xf from Vd infection caused by the two xylem-limited pathogens that trigger similar visual symptoms.

Automated summary

A three-stage machine learning algorithm was proposed to distinguish Vd infections from Xf infections, using datasets from 27 olive orchards affected by Xf and Vd outbreaks in Italy and Spain between 2011 and 2017. Plant traits derived from airborne hyperspectral and thermal imagery successfully discriminated Xf from Vd infections. The method generated false-positive rates of 9% for Vd infections and 4% for Xf infections, with overall accuracies of 98% and 92%, respectively.