A. Hornero; R. Hernández-Clemente; P.R.J. North; P.S.A. Beck; D. Boscia; J.A. Navas-Cortes; P.J. Zarco-Tejada.
November 18, 2019
Outbreaks of Xylella fastidiosa (Xf) in Europe generate considerable economic and environmental damage, and this plant pest continues to spread. Detecting and monitoring the spatio-temporal dynamics of the disease symptoms caused by Xf at a large scale is key to curtailing its expansion and mitigating its impacts. Here, we combined 3-D radiative transfer modelling (3D-RTM), which accounts for the seasonal background variations, with passive optical satellite data to assess the spatio-temporal dynamics of Xf infections in olive orchards. We developed a 3D-RTM approach to predict Xf infection incidence in olive orchards, integrating airborne hyper- spectral imagery and freely available Sentinel-2 satellite data with radiative transfer modelling and field ob- servations. Sentinel-2A time series data collected over a two-year period were used to assess the temporal trends in Xf-infected olive orchards in the Apulia region of southern Italy.
Hyperspectral images spanning the same two- year period were used for validation, along with field surveys; their high resolution also enabled the extraction of soil spectrum variations required by the 3D-RTM to account for canopy background effect. Temporal changes were validated with more than 3000 trees from 16 orchards covering a range of disease severity (DS) and disease incidence (DI) levels. Among the wide range of structural and physiological vegetation indices evaluated from Sentinel-2 imagery, the temporal variation of the Atmospherically Resistant Vegetation Index (ARVI) and Optimized Soil-Adjusted Vegetation Index (OSAVI) showed superior performance for DS and DI estimation (r2VALUES > 0.7, p < 0.001). When seasonal understory changes were accounted for using modelling methods, the error of DI prediction was reduced 3-fold. Thus, we conclude that the retrieval of DI through model inversion and Sentinel-2 imagery can form the basis for operational vegetation damage monitoring worldwide. Our study highlight the value of interpreting temporal variations in model retrievals to detect anomalies in vegetation health.
Full paper via Science Direct [.pdf | 7,4 MB]