Monitoring tar spot disease at different canopy and temporal levels using aerial multispectral imaging and machine learning

Authors: ZHANG, CHONGYUAN - Purdue University; LANE, BRENDEN - Purdue University; FERNANDEZ-CAMPOS, MARIELA - Purdue University; CRUZ-SANCAN, ANDRES - Purdue University; LEE, DA-YOUNG - Purdue University; GONGORA-CANUL, CARLOS - Purdue University; ROSS, TIFFANA - Purdue University; DA SILVA, CAMILA - Purdue University; TELENKO, DARCY - Purdue University; Goodwin, Stephen - Steve; Scofield, Steven - Steve; OH, SUNGCHAN - Purdue University; JUNG, JINHA - Purdue University; CRUZ, C.D. - Purdue University

Submitted to: American Phytopathological Society Abstracts
Publication Type: Abstract Only
Publication Acceptance Date: 5/5/2023
Publication Date: 8/12/2023
Citation: Zhang, C., Lane, B., Fernandez-Campos, M., Cruz-Sancan, A., Lee, D., Gongora-Canul, C., Ross, T.J., Da Silva, C.R., Telenko, D.E., Goodwin, S.B., Scofield, S.R., Oh, S., Jung, J., Cruz, C. 2023. Monitoring tar spot disease at different canopy and temporal levels using aerial multispectral imaging and machine learning. American Phytopathological Society Abstracts. ABSTRACT.

Interpretive Summary: N/A

Technical Abstract: Tar spot is a high-profile disease, causing yield losses (up to 58%) on corn (Zea mays L.) in several countries throughout the Americas. Tar spot symptoms usually appear at the lower canopy in corn fields with infection history, making disease monitoring through unmanned aircraft systems (UAS)-based imaging alone very challenging due to occlusion of the upper canopy. UAS-based multispectral imaging and machine learning were integrated to monitor tar spot at different canopy and temporal levels and evaluate the early onset of disease and efficacy of multiple disease management tactics. Disease severity was assessed visually at three canopy levels within micro-plots, while aerial images of corn fields were acquired by UASs equipped with multispectral cameras. Both disease severity and multispectral images were collected from five to eleven time points each year for two years. Image-based features, such as vegetation indices (VIs) and their statistics, were extracted from ortho-mosaic images and used as machine learning inputs to develop disease quantification models. The developed quantification models showed encouraging performance in estimating disease severity at different canopy levels in both years (coefficient of determination of 0.75 to 0.93 in most cases and Lin's concordance correlation coefficient of 0.75 to 0.97). Epidemiological parameters of temporal disease progress, including initial disease severity or y0 and area under the disease progress curve, were successfully modeled using estimated disease severity. Results illustrated that the extracted disease epidemiological information could be used to monitor the onset of tar spot even though disease severity is relatively low (< 1%) and to evaluate the efficacy of disease management tactics under micro-plot conditions (Spearman’s rank correlations between the actual and estimated rankings of 0.54 to 0.97). Further studies are required to apply and validate our digital phenotyping-based tar spot disease monitoring methods to large corn fields and associated value in supporting decision-making.