A Case Study for Visual Detection of a Systemic Disease: Optimizing Identification of Phony Peach Disease Based on Statistical and Machine Learning Models

Phony peach disease (PPD), caused by <i>Xylella fastidiosa</i> subsp. <i>multiplex</i> (Xfm), poses a significant threat to commercial peach orchards in Georgia. Early and accurate detection is essential for effective disease management, yet visual assessment remains the primary approach for diagnosing PPD symptoms due to the high cost and logistical challenges of qPCR-based detection of Xfm. We evaluated the accuracy of visual PPD assessment and examined the factors influencing rater performance, symptom reliability, and optimal survey deployment strategies with classification and regression trees/random forest analyses and simulations. Internode length was the most reliable symptom for PPD identification in two peach cultivars, consistently outperforming other physical traits, such as canopy flatness and shape. Primer pair C06Xf-bamA had the greatest relative sensitivity, making it the preferred choice for qPCR confirmation. Principal component analysis suggested that rater experience significantly improved agreement with qPCR results, and repeated assessments of the same orchards further enhanced consistency for raters. Simulation results suggested that deploying two experienced raters may provide the highest detection diagnostic accuracy for survey purposes, particularly when qPCR-based pathogen detection is unavailable. Finally, PPD-affected trees, through PCR verification and visual identification, exhibited higher mortality rates than Xfm-negative trees, reinforcing the need for early detection and removal to limit disease spread. These findings underscore the importance of strategic rater deployment, targeted symptom selection, and integrating molecular diagnostics when feasible.