Alternaria Leaf Blight and Head Rot of Broccoli: UAV-Based Disease Detection and Fungicide Resistance Management

Abstract

Alternaria leaf blight and head rot (ABHR) of brassicas significantly impacts yield, threatening the sustainability of the broccoli industry. Management traditionally relies on quinone outside inhibitor (QoI) fungicides, but recent reports indicate reduced efficacy. The causes for this decline are unclear, whether due to a novel species or resistance within existing species. Additionally, knowledge gaps exist regarding disease-tolerant broccoli cultivars, the extent of QoI fungicide compromise in Virginia, and effective alternative control measures. Disease monitoring is crucial for timely detection of pathogens and fungicide program failures. Multispectral imaging is increasingly adopted for disease monitoring in specialty crops, though its efficacy for ABHR in broccoli fields is unexplored. During the 2021-2024 field seasons, samples with ABHR symptoms were collected from brassica crops and weeds across Virginia. Alternaria species were identified via PCR using genus- and species-specific primers (ALT, ABRA, ABRE, AJAP) and molecular identification through ITS, TEF1, Alt a1, and RPB2 genes. A. brassicola and A. alternata were identified as the predominant species in the region. Pathogenicity varied by region and host. Field trials assessed ABHR susceptibility of 26 broccoli cultivars and the effectiveness of fungicides. Disease severity and yield varied by year. 'Vallejo', 'Marathon', and 'Belstar' had the highest yields, while 'Eastern Crown', 'Green Magic', and 'Millennium' had the lowest. Biological fungicides (Oxidate 5.0, OSO, Guarda) and conventional fungicides (Topguard, Inspire Super, Luna Sensation) effectively reduced disease severity. Marketable yield was highest with Miravis Prime, Fontelis, and Topguard. A UAV-mounted five-band multispectral camera performed imaging across three broccoli fields. Visual assessments of disease severity preceded imaging. Severity was determined using five spectral bands and 74 vegetation indices (VIs) from 184 observations. Multivariate analyses identified the best predictors of disease severity. K-means clustering and PCA classified disease severity into three levels with a Silhouette Score of 0.34. Among regression models, Random Forest achieved the highest performance (MSE = 44.45, R² = 0.79), while classification models such as Random Forest, Artificial Neural Networks, and Gradient Boosting attained 88, 86, and 85% accuracy respectively. Feature importance analysis highlighted indices like TCARI, CVI, and SRPI as consistently influential across models. These findings demonstrate the potential of combining multispectral imaging with machine learning to enhance early detection and management of ABHR in broccoli crops.