Detecting X. fastidiosa with hyperspectral remote sensing: findings from two years of airborne campaigns in Puglia
A remote sensing campaign carried out in Puglia in summer 2016 collected hyperspectral and thermal images of ca 200,000 olive trees at sub-meter resolution. In the 1200 ha study area within the X. fastidiosa infected zone, 3500 trees were simultaneously evaluated in the field for severity of X. fastidiosa symptoms. The hyperspectral sensor used in this experiment acquired data in the visible and near-infrared spectral region (400-885 nm) with 260 bands of 6.5 nm FWHM at 1.85 nm/pixel and 12-bit radiometric resolution. The sensor was radiometrically calibrated in the laboratory, and images atmospherically corrected to obtain surface reflectance using total incoming irradiance measured in the field. The high resolution hyperspectral and thermal imagery acquired over the orchards allowed the delineation of individual trees using object-based crown detection algorithms. Crown temperature and hyperspectral indices were calculated for each tree to classify disease severity levels using different machine learning algorithms, including linear discriminant analysis, support vector machines and neural networks. The success and applicability of these early detection methods to other areas will be discussed in the context of a new airborne campaign planned in July 2017 over X. fastidiosa infected zones in Mallorca. In this new airborne campaign the impact of X. fastidiosa symptoms on crown reflectance will also be evaluated in the 800-1700 nm spectral region. Acknowledgment This work was supported by JRC and has received funding from from the European Union's Horizon 2020 research and innovation programme under grant agreement N. 635646 "Pest Organisms Threatening Europe POnTE" and grant agreement N. 727987 "Xylella fastidiosa Active Containment Through a multidisciplinary-Oriented Research Strategy XF-ACTORS".