José Ramón Torres Martín

Xylella fastidiosa (Xf) is a vascular plant pathogen that affects several crops by blocking xylem vessels, impairing water and nutrient flow, and ultimately leading a plant decline. Early detection is challenging due to symptom latency and overlap with abiotic stress. This study presents a methodological framework for temporal monitoring of Xf infection using leaf-level physiological and spectral traits. Data were collected from up to two selected branches of each of 96 almond trees across four dates in 2024. At each date, eight leaves per branch were measured and discarded after their analysis, enabling temporal monitoring at the branch level. Leaf samples were classified into four categories based on molecular tests, visual symptoms, and expert assessments, considering whether the corresponding branch was already infected or became symptomatic during the study period. Non-parametric tests identified informative variables for Xf -detection, including pigment and structure related indices (e.g. TCARI, MCARI) and physiological traits (e.g. stomatal conductance, fluorescence). A simple and interpretable data-driven predictive model, in this work a logistic regressor, was trained using the identified variables on leaf-level data from trees with contrasting health conditions (healthy vs. symptomatic). The model was evaluated for its ability to generalize over time. Models trained with data from the first two measurement dates showed the best generalization, achieving weighted accuracies between 60 % and 80 % even in later dates, and improved detection of symptomatic branches. Data from the second and third dates offered the best separation between health conditions, while the last date was less informative due to confounding stress factors. Despite limitations from sample size and class imbalance, the framework demonstrates potential for non-invasive disease monitoring in woody crops.