Leaf Diffuse Reflectance Spectroscopy for Early Detection of Ceratocystis Wilt in Eucalyptus Cuttings

The increasing global demand for products derived from Eucalyptus spp. has stimulated its production in Brazil. However, productivity has declined in recent years due to multiple factors, with Ceratocystis wilt among the major causes. Conventional detection methods rely on visual assessment, histological sections, and/or molecular analyses—procedures that are time-consuming and impractical at scale. Proximal or remote sensing based on VIS–NIR–SWIR spectroscopy (400–2500 nm) has been proposed as non-destructive alternatives for characterizing plant biochemical and biophysical properties, yet its use for detecting Ceratocystis wilt in Eucalyptus spp. remains underexplored. Here, we evaluated whether leaf reflectance measurements in the VIS–NIR–SWIR, acquired with a proximal non-imaging sensor, can be used to detect the disease in asymptomatic cuttings (vegetatively propagated plants). For that a greenhouse experiment was established with two Eucalyptus clones, one susceptible and another resistant. Plants were visually assessed and tested via the “carrot bait” method for disease incidence, and spectral measurements collected four times between 12 and 60 hours after inoculation. Observations for inoculated plants were compared with those from non-inoculated references (total n = 77). Classification models trained with Partial Least Squares Discriminant Analysis (PLS-DA), Random Forest with Recursive Feature Elimination (RF+RFE), and Support Vector Machine with a Genetic Algorithm (SVM+GA) achieved balanced accuracy of 0.63 ± 0.11, 0.75 ± 0.11, and 0.75 ± 0.13, respectively. Features selected via RFE and GA, or identified as highly important in the PLS-DA, RF+RFE, and SVM+GA models, were mainly located within the visible, NIR, and particularly the SWIR regions. This distribution is consistent with absorption features associated with leaf water, cellulose, starch, and lignin (near 1100–1200 and 2300 nm), as well as proteins (near 1700, 2200, and 2300 nm). Spectra from the apical canopy layer generally provided better classification performance than from the basal or middle canopy sections. Despite the relatively small dataset and limited number of clones, our results demonstrate the potential of proximal spectroscopy for detecting Ceratocystis wilt in asymptomatic Eucalyptus plants.