Early Detection of Ceratocystis Wilt in Eucalyptus through Leaf Diffuse Reflectance

Ceratocystis wilt caused by Ceratocystis fimbriata is an important disease of Eucalyptus. Proximal or remote sensing methods based on Vis-NIR-WIR spectroscopy (400 and 2500 nm) have been proposed as a suitable non-destructive alternative to characterize plant biochemical and biophysical properties in different contexts. In this study, the potential for using leaf reflectance in the Vis-NIR-SWIR to detect this disease in asymptomatic seedlings was evaluated in a greenhouse experiment with two eucalyptus clones (ref. nbr. 223 – susceptible and ref. nbr. 236 – resistant). Plants were evaluated between 1 and 60 hours after inoculation and compared to non-inoculated references (n = 97 plants). Average spectral differences between inoculated and non-inoculated plants were found before visual symptoms appeared for both clones. The apical leaf layer showed the largest spectral differences before and after visual symptoms, in comparison to the canopy basal and middle leaf layers. Random Forest models coupled with Recursive Feature Elimination (RFE) resulted in balanced accuracy of 0.596 for the validation dataset. Using two band Vegetation Indices (VIs) optimized specifically for Ceratocystis wilt detection allowed slightly better performance, resulting in balanced accuracy of 0.644 for the validation dataset. Features selected via RFE or via VIs belonged mostly to NIR and specially to SWIR wavelengths, which can be associated to changes in leaf water, cellulose, starch and lignin contents (near 1100-1200 and 2300 nm) and proteins (near 1700, 2200 and 2300 nm). Our results suggest potential for detecting Ceratocystis wilt in Eucalyptus spp. using proximal or remote spectroscopy in asymptomatic plants.