Scalable Early Detection of Grapevine Viral Infection with Airborne Imaging Spectroscopy
Scalable Early Detection of Grapevine Viral Infection with Airborne Imaging Spectroscopy
复制标题
利用机载成像光谱对葡萄病毒感染进行可扩展的早期检测
DOI：
10.1094/phyto-01-23-0030-r
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Dokoozlian, Nick
中科院分区：
文献类型：
--
作者：
Galvan, Fernando E.;Pavlick, Ryan;Trolley, Graham;Aggarwal, Somil;Sousa, Daniel;Starr, Charles;Forrestel, Elisabeth;Bolton, Stephanie;Alsina, Maria del;Dokoozlian, Nick
美国葡萄酒和葡萄产业每年因病毒性疾病损失30亿美元，包括葡萄藤叶病毒复合体3（GLRaV-3）。目前的检测方法是劳动密集型和昂贵的。GLRaV-3具有潜伏期，在此期间葡萄藤被感染但不显示可见症状，使其成为评估基于成像光谱的疾病检测的可扩展性的理想模型。NASA下一代机载可见光和红外成像光谱仪于2020年9月部署在加利福尼亚州洛迪的赤霞珠葡萄藤中检测GLRaV-3。在图像采集后不久，作为机械收获的一部分，从葡萄藤上除去叶子。于二零二零年及二零二一年九月，行业合作者逐棵葡萄树搜寻317公顷土地，以寻找可见病毒症状，并收集部分样本进行分子确认测试。假设2021年发现的症状葡萄藤在图像采集时已被潜伏感染。随机森林模型在未感染和GLRaV-3感染的葡萄藤的光谱信号上进行训练，该光谱信号与未感染和GLRaV-3感染的葡萄藤的合成少数过采样相平衡。该模型能够区分未感染和GLRaV-3感染的葡萄树，无论是前和posteradically在1至5米的分辨率。表现最好的模型在区分未感染和无症状葡萄树之间的准确率为87%，在区分未感染和无症状+有症状葡萄树之间的准确率为85%。不可见波长的重要性表明，这种能力是由疾病引起的植物生理变化驱动的。这些结果为使用即将到来的高光谱卫星表面生物学和地质学用于葡萄藤和其他作物物种的区域疾病监测奠定了基础。版权所有© 2023作者。这是一篇开放获取的文章，在CC BY-NC-ND 4.0国际许可下分发。
The U.S. wine and grape industry loses $3B annually due to viral diseases including grapevine leafroll-associated virus complex 3 (GLRaV-3). Current detection methods are labor-intensive and expensive. GLRaV-3 has a latent period in which the vines are infected but do not display visible symptoms, making it an ideal model to evaluate the scalability of imaging spectroscopy-based disease detection. The NASA Airborne Visible and Infrared Imaging Spectrometer Next Generation was deployed to detect GLRaV-3 in Cabernet Sauvignon grapevines in Lodi, CA in September 2020. Foliage was removed from the vines as part of mechanical harvest soon after image acquisition. In September of both 2020 and 2021, industry collaborators scouted 317 hectares on a vine-by-vine basis for visible viral symptoms and collected a subset for molecular confirmation testing. Symptomatic grapevines identified in 2021 were assumed to have been latently infected at the time of image acquisition. Random forest models were trained on a spectroscopic signal of noninfected and GLRaV-3 infected grapevines balanced with synthetic minority oversampling of noninfected and GLRaV-3 infected grapevines. The models were able to differentiate between noninfected and GLRaV-3 infected vines both pre- and postsymptomatically at 1 to 5 m resolution. The best-performing models had 87% accuracy distinguishing between noninfected and asymptomatic vines, and 85% accuracy distinguishing between noninfected and asymptomatic + symptomatic vines. The importance of nonvisible wavelengths suggests that this capacity is driven by disease-induced changes to plant physiology. The results lay a foundation for using the forthcoming hyperspectral satellite Surface Biology and Geology for regional disease monitoring in grapevine and other crop species.Copyright © 2023 The Author(s). This is an open access article distributed under the CC BY-NC-ND 4.0 International license.